---
id: "workshop:introducing-grantwell-using-ai-for-municipal-technical-assistance"
type: "InnovateUS workshop"
title: "Introducing Grantwell: Using AI for Municipal Technical Assistance"
url: "https://innovate-us.org/introducing-grantwell-using-ai-for-municipal-technical-assistance"
date: "2026-02-25"
duration: "60 minutes"
language: "English"
series: "Ideas in Action"
instructors: ["Quentin Palfrey","Anjith Prakash Chathan Kandy"]
recording: "https://www.youtube.com/watch?v=IUbjzmumIp0"
transcript: "automatic, from the recording"
source: "InnovateUS (innovate-us.org)"
licence: "CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org"
---

# Introducing Grantwell: Using AI for Municipal Technical Assistance

InnovateUS workshop · 25 February 2026 · 60 minutes

In this session, Quentin Palfrey, Director of the Office of Federal Funds for the Commonwealth of Massachusetts, and Anjith Prakash Chathan Kandy, AI engineering at The Burnes Center for Social Change, will introduce Grantwell, an AI-enabled grant-writing assistance tool developed by the Massachusetts Federal Funds and Infrastructure Office in partnership with the Northeastern Burnes Center for Social Change. They will explain why the tool was created, how it works, and where it fits into real municipal workflows. Participants will explore how AI can support (not replace) core grant-writing tasks, such as researching funding opportunities, structuring proposals, and strengthening clarity, while maintaining accuracy, transparency, and public accountability.

This workshop is designed for municipal and local government professionals looking for practical, responsible ways to use AI to move grant applications forward and improve outcomes for their communities.

## Learning goals

By the end of this workshop, participants will be able to:

- **Identify** where AI can support key stages of the grant-writing process, including opportunity scanning, outlining proposals, and refining key sections

- **Apply** responsible AI practices to maintain accuracy, confidentiality, compliance, and equity when generating or editing grant content

- **Use** AI tools to strengthen the clarity, structure, and persuasiveness of grant narratives while ensuring that human expertise and community priorities guide final decisions

## Speakers

- **Quentin Palfrey**, Director of Federal Funds and Infrastructure Office of Massachusetts Governor Maura Healey
- **Anjith Prakash Chathan Kandy**, AI for Impact Fellow, The Burnes Center for Social Change

Part of the series [Ideas in Action](https://innovate-us.org/workshop-series/ideas-in-action).

Recording: https://www.youtube.com/watch?v=IUbjzmumIp0

## Transcript

Automatic transcript of the recording: names and terms can be misheard, and the times can be a few seconds off. Each time links to that moment of the video. The transcript alone: https://innovate-us.org/introducing-grantwell-using-ai-for-municipal-technical-assistance.txt (plain text) and https://innovate-us.org/introducing-grantwell-using-ai-for-municipal-technical-assistance.json (every segment with its start and end second).

**Quentin Palfrey**

[0:00](https://youtu.be/IUbjzmumIp0?t=0) Terrific. Thank you so much. And it's great to be here. My name is Quentin Palfrey. I'm the Director of Federal Funds and Infrastructure in Massachusetts. And I work for a governor, Maura Healey, who came into office in 2023. And as you may recall, at that time, Congress had recently passed three pieces of major legislation that collectively represented the largest investment in infrastructure in the history of the country. One of those pieces of legislation was called the Bipartisan Infrastructure Law. Another, the Inflation Reduction Act, actually represents the largest climate bill in the history of the world. And the Chips and Science Law. And one of the things that Governor Healey committed to as a day one commitment was that we were going to be aggressive and systematic in pursuing these resources and bringing them back into Massachusetts to advance a number of priorities across the Commonwealth. And the way that these laws are set up is relatively complicated. Some of the resources came into the state governments directly. Some of the resources were available to cities, towns, tribes, and other organizations who then needed to navigate the complexity of the federal government. And then some things worked their way into communities through the tax code and through other mechanisms.

[2:02](https://youtu.be/IUbjzmumIp0?t=122) And so what we did was we created a whole-of-government effort. The Governor created a new position that reported to her a new office within our Executive Office of Administration and Finance. And then a series of mechanisms to make sure that we were working together to advance these priorities, including an interagency task force, a lot of work with our federal partners. But one piece of this was particularly important to us, which was making sure that cities, towns, and tribes were able to have the resources and information both to know about these opportunities and to effectively pursue them. And so in order to advance that part of our charge, we launched something that we call the Massachusetts Federal Funds Partnership. And this is essentially a platform for the state government to interact with a number of cities, towns, and tribes, help them understand what resources the federal government is offering, help them understand the programs within the state government, and help them to pursue those opportunities. If you move to the next slide, one of the things that we thought was really important was for us to hear directly from communities about what their needs were. So we spent a lot of time going out into communities across Massachusetts, getting a sense of how they were experiencing these new federal programs, getting a sense of what resources they wanted us to provide in order to support them in moving forward infrastructure projects. And we also enlisted the aid of the Harvard Kennedy School, who dedicated what they called a field lab to our work and helped us to do some analysis of

[4:10](https://youtu.be/IUbjzmumIp0?t=250) the needs. If you could move to the next slide. And we heard pretty consistently the same kinds of challenges. So we heard that some folks felt that the programs were very complex. It was very hard to get good information about what folks could apply for. The timelines, the deadlines, the federal silos were very complicated. There was a number of barriers that we heard that people faced. Some of them were it was very hard to do the early planning work for important projects. It was hard to do the grant writing and the implementation of the work. It was hard to come up with the resources. So in many cases, the federal programs require the communities to put some money into these projects to match the federal dollars that they're seeking. And we heard a lot of challenges that were particularly unique to smaller and under-resourced communities. So ironically, although some of these major federal programs were set up in a way that prioritized rural communities, underserved communities, environmental justice communities, ironically, a lot of these communities were still having difficulty accessing those resources.

[5:47](https://youtu.be/IUbjzmumIp0?t=347) That wealthy communities often had consultants. They often had large planning departments. They had track records of being able to pursue these resources. And rural communities, underserved communities, the very communities that some of these programs were designed to try to particularly help were sometimes the communities that faced the greatest barriers to entry. And so as we moved forward with implementing the governor's vision on how to leverage these new resources, we put in place a number of programs that sought to address the gaps that we heard when we engaged with cities, towns, and tribes. And we wanted to fit the solutions to the challenges that we knew from our community engagement that communities were facing. So one of the things that we were able to do, so we were able to implement some programs right away through our existing state resources, but we also asked the legislature to work with us to put new resources on the table. And we used a somewhat unusual mechanism for this. So Massachusetts has a rainy day fund, a stabilization fund, that over the last couple of decades has become relatively robust. And we see that rainy day funds can be really helpful in navigating difficult times. So we didn't ask the legislature to be able to tap into the corpus of that. But we said, you know, if you could let us take the interest on the rainy day fund, put it into a separate pool,

[7:48](https://youtu.be/IUbjzmumIp0?t=468) we can use this to entice the federal government to give us new resources and to support projects that can be supercharged if both federal resources and this money from the interest on the rainy day fund could come into it. And the legislature, with broad bipartisan support, with unanimous support in both houses of the legislature, with the business community, municipalities, labor, environmental groups, it was really broad-based coalition support, passed a piece of legislation, gave us $750 million, which for a state of our size is a huge amount of resources for this program, to do a series of things. Some of it was to match these applications for state projects, clean energy, economic development, technology, transportation, but also to support cities, towns, tribes, regional planning agencies as they pursue these tools. We also asked them, in sync with some of what we heard from our community partners, to give us some new resources to provide technical assistance to communities. So as communities were trying to plan these projects, do design and engineering, grant writing, implementation, we wanted to be able to give them new resources. And we also asked for new lending tools so we could put nimble, low-interest loans in to support some of these projects. So the project that we're talking about here today emerges from that really deep conversation with cities, towns, and tribes across Massachusetts.

[9:34](https://youtu.be/IUbjzmumIp0?t=574) Because one of the barriers that we saw was that smaller communities, underserved communities, were finding it really hard to get going with grant applications, to find the materials, to understand what the requirements were, to understand the deadlines. Now we knew that just giving them direct support wasn't going to work, and so we carved off some resources to give some grants for some of these communities to hire staff or consultants or to do other design engineering work. And yesterday we announced the first 11 recipients of our municipal and tribal technical assistance program, and we announced that all of the regional planning agencies in Massachusetts are receiving awards in connection with this program to support communities. So the GrantWell program that we're going to describe in a moment is a piece of a larger strategy to support cities, towns, and tribes as they navigate this environment of trying to finance and move forward important infrastructure projects. But there are other pieces to that strategy as well. And before I get into sort of how GrantWell fits into this puzzle, I also need to acknowledge that the landscape of federal resources and of federal-state relations has changed really dramatically since we launched this effort. So in 2023, we were living in a world where there were new programs and there was an abundance of competitive discretionary grant programs at the federal level. We are now in a world where there have been a lot of disruptions to those programs, a lot of challenges that communities and other partners are facing. And so we've adapted to some extent our strategy as the circumstances have changed to incorporate

[11:36](https://youtu.be/IUbjzmumIp0?t=696) trying to help people find resources from federal sources, but also to be able to navigate the complex state resources and charitable and foundation resources that may help them to move their projects forward. But we actually find that this strategy of giving tools, new tools, to communities to identify the resources that are out there to move projects forward is actually more important in the landscape that we find ourselves in than it even was when we designed it. So if you move to the next slide, one of the, I'm sorry, this slide just covers one other piece of this program, which is that in addition to helping folks pursue grants, we are also developing new lending resources for nimble, low-interest loans to support communities where grants don't fit or where they need to be supplemented.

[12:38](https://youtu.be/IUbjzmumIp0?t=758) But if you move to the next slide, but we think that there's a real, really important role for artificial intelligence in this strategy. So we are not in a position as a medium-sized office, even as a relatively large state government network, to roll up our sleeves and actually perform the grant writing function for all of our community partners. And there are many ways in which we can support through this new grant program and through our regional planning agencies providing some additional resources. But there are a lot of circumstances where community partners are in a situation where they need to figure out what are the grants that are available to meet their need? What are the requirements of those grants? And how do they begin to populate a grant application that is very likely to be successful?

[13:48](https://youtu.be/IUbjzmumIp0?t=828) Massachusetts has been very fortunate to have a partnership with an academic partner at Boston University. And the Byrne Center for Social Change has been this really wonderful partner for the Commonwealth of Massachusetts. And we've worked with them on an array of challenges. But the Byrne Center has been with us every step of the way in developing this new artificial intelligence tool. And when I say every step of the way, the tool was actually developed in part through our roadshow. So as we went out to different communities and asked them what do they need? What are their challenges? How can we help? What grants are you applying for? We brought the Byrne Center along with us. And they were able to take feedback from precisely the community leaders that we were developing this tool to try to assist. And so the product that Anjith is going to sort of walk you through in a moment is the product of that Collaboration. What Grantwell is at base is something that we're making available to the public as an artificial intelligence-powered tool to help communities who are seeking and applying for funding opportunities. The first base case was the federal discretionary grant opportunities, but it is something that can be adapted and expanded and that we're adapting and expanding to include state applications, state grant programs, as well as charitable grant programs. And we see this as an important tool not only in helping communities build projects or programs that support community needs, but also an important tool

[15:53](https://youtu.be/IUbjzmumIp0?t=953) in overcoming an equity gap, a gap that we see underserved and rural communities facing as they navigate complicated programs. So with that, what I'd love to do is introduce my colleague, Anjit, who can walk through some of this terrific new tool that the Burns Center has developed

**Anjith Prakash**

[16:14](https://youtu.be/IUbjzmumIp0?t=974) in partnership with our office. Anjit. Thank you, Quinton. I'll just share my screen. Yep. So hey, my name is Anjit. I'm an AI for Impact fellow at the Burns Center at Northeastern University. So as he mentioned, Grantwell was born through the partnership between the Federal Funds and Infrastructure Office and the Burns Center AI for Impact team. So when we first started, we assumed that the main issue was the grant writing itself. So naturally, we thought, let's build an AI tool to help write grants. That should solve it, right? But once we got into the field, we realized that the problem was bigger than just writing. So with the first version of the tool, we had the opportunity to join the FFIO team on the statewide roadshows across Massachusetts.

[17:20](https://youtu.be/IUbjzmumIp0?t=1040) So this was a turning point for us. In those sessions, we sat with local officials, municipal leaders, and community organizations. We listened. So we heard directly from the real barriers they faced when trying to access the federal funding, limited staff capacity, and also the complex requirements, long, no-fault documents, and the answer. So from those conversations, three clear things were emerged, things like finding the right grants, understanding the complex requirements, and writing strong competitive proposals. It wasn't just about the drafting, but it was more about navigating the entire process. So that's where we realized AI could actually make a meaningful impact, not by replacing people, but by reducing the friction. So we evolved Granville.

[18:21](https://youtu.be/IUbjzmumIp0?t=1101) Instead of throwing AI at the problem, which our director always says is a very bad idea, we focused on solving the real pain points communities described. So Granville is an AI powered platform designed to help underserved municipalities pursue the federal funding more efficiently. So our goal is very simple. Simplify the grant process so that every city and town, regardless of their size or internal capacity, has a fair shot at competing. So based on this user feedback we got from the roadshows, we built the no-fault summarization, an interactive chat bot, and a fully guided Granville grant writing tool. So I'll go through each of the sections. So this was the features what I said earlier. So I'll give you a quick walkthrough of the tool so that is it all visible, all good?

[19:24](https://youtu.be/IUbjzmumIp0?t=1164) Okay, perfect. Yes. So thank you. So here's a quick walkthrough of the tool. So at the foundation, we built an automated system that pulls the federal funding opportunities directly from the public sources and ingests them into a database. So these public sources are like grants.gov. And instead of these communities across multiple websites, we centralize these structures in one place. So from here, users can browse and select a relevant NOFO. As you can see here, right now the users typically begin by choosing a known grant. But in reality, many communities don't start with a grant. They start with a need. For example, we need to upgrade our wastewater system or we are trying to expand rural broadband or something like that. So we are developing a feature where users can simply describe their use case in plain language. So instead of searching by grant name or agency, they can just describe their project goals, the scope or the impact. So from there, Granville looks into the use case, matches it against our full grants database to get the most relevant funding opportunity. So instead of asking, is this grant right for me? It's more like given the project, which grants are the best. So community students have to decode the federal funding ecosystem just to know where they are fit. They just simply describe what they're trying to build and the system does the mapping. So once the NOFO is selected, I'll just select one.

[21:16](https://youtu.be/IUbjzmumIp0?t=1276) So once the NOFO is selected, the summarization tool breaks down what is often a 40 to 50 page NOFO document, extracts the most important sections, which are the eligibility criteria, the required documents, the narrative sessions, and the key deadlines. So this simply shows if they are eligible for this. For example, I'm a city from Quincy, for example. I could be like, hey, am I eligible for this? Or the documents required to submit the grant application, which all sections are to be written for the grant and the deadlines. So this is pretty funny how we got into deadlines because when we spoke to these grant writers, they were like, it's pretty hard to keep track of the several deadlines. There are like a million dates in the NOFO document. So it's pretty hard. So that's where we came to this. So this allows the people to understand a grant is a good fit before investing significant time. Next, we have a chat bot, which users can ask any clarifying questions grounded directly in the source document.

[22:30](https://youtu.be/IUbjzmumIp0?t=1350) For example, what is the cost share of this grant? So the response is based only on the NOFO content. So users can quickly navigate the complex requirements without manually scanning through the dozens of pages. So here it says it's 80% which federal funds would be giving up and then the 20 should be a cost share which we have to put in ourselves. So now we are moving into the grant writing tool. Here what we are trying to do is we have split the sections into project basics and questionnaire. So what happens here is users can enter their project specific details with the selected NOFO. The system guides the users through a structured questionnaire which is specifically aligned to the NOFO selected and which generates a tailored draft. So I have kept a pre-filled section here so that it's easier for us to do it. So I've just filled up the details project name, the requested amount, and then this questionnaire what we have seen here, it is very specific to the specific NOFO so that it would be easier for them to find out.

[23:57](https://youtu.be/IUbjzmumIp0?t=1437) So it's not generic ones as a normal tool would give which is very specific to this grant. So once you have entered all the answers, they have also a section where you can also add any supporting documents. You can maybe upload your previous grants, any statement of support, and that would support your case for the draft and if there is any additional information as well. So in the interest of time, I have already done a part of this so I'll just load that up. Just give me a second. Yep, yep. So in the interest of time, I have just loaded up the draft which was already generated.

[25:16](https://youtu.be/IUbjzmumIp0?t=1516) So here's the generated draft. So here you can see that all these sections are written. Can you see the screen? Yeah, okay. So this is the generated draft. What I could say here is this doesn't replace the human judgment. It gives teams a strong starting point, especially for the municipalities with limited staffing capacity. So finally, what users can do is they can review, edit, and refine and export the application as a PDF. So you can just write up here. So with Granville, communities can reduce the time spent navigating through complex federal documents, increase the submission readiness, and improve access to federal dollars. And ultimately, this matters because smaller municipalities deserve the same opportunity to compete for funding as larger, better resource cities.

[26:25](https://youtu.be/IUbjzmumIp0?t=1585) And Granville is all about leveling that playing field. Thank you so much.

**Agueda Quiroga**

[26:30](https://youtu.be/IUbjzmumIp0?t=1590) Okay, thank you so much. I am Skalin again. Okay, it's Quentin again. So we are totally open for questions. There's first question in the Q&A. I don't know, I am going to read it and just decide who is going to answer. It's from Sybil and it says, does this interface also serve as a project management platform?

**Quentin Palfrey**

[27:08](https://youtu.be/IUbjzmumIp0?t=1628) Anjeet, why don't you take that one?

**Anjith Prakash**

[27:13](https://youtu.be/IUbjzmumIp0?t=1633) No, so for project, it's not a project management. It's more like we would be handling the draft until the first draft. So once the first draft is written, it can be taken down as a Microsoft Word document, and then you can add on to it. So the inspiration why we did this is when, so there could be a city where they could be like one person doing like five to 10 jobs. So for them, they find it difficult to start the draft rather than write the rest of it. So this serves as like a starting point where we give them a structured output where they can just

**Quentin Palfrey**

[27:55](https://youtu.be/IUbjzmumIp0?t=1675) develop according to their needs. Yeah, we're trying to take some of the friction out of the grant application process. It's really important for the applicant to be the final, to own the materials and to really review and to adjust. But one of the things that we thought that this tool could do was help get people to the place where they had a really high quality first draft and help them to navigate around the program requirements so that they had something to work with. So it doesn't replace the human element, but it can make the earlier phase of the grant application process a lot easier, particularly for our under-resourced communities.

**Agueda Quiroga**

[28:41](https://youtu.be/IUbjzmumIp0?t=1721) Great. There's another question by Heather. Is the tool ready for states other than Massachusetts to

**Quentin Palfrey**

[28:47](https://youtu.be/IUbjzmumIp0?t=1727) utilize? I'll give that one to Antith. But our vision on this is absolutely for this to be something that can be shared more broadly. It also can be adapted. And so one of the things that we've started to do is really try to identify some local grant opportunities. We know that there are silos within state government as well as within the federal government. And so we've started to kind of seek out some of the geography-specific opportunities. We've also started to think about how to partner with some of our philanthropic communities. They put out various kinds of resources. So if you take, say, the climate space, for example, this is a space where changes in federal government policy have been very disruptive. And in some ways, the philanthropic community has tried to offer some new resources to solve some of the challenges in environmental justice and resilience and clean energy and decarbonization.

[29:48](https://youtu.be/IUbjzmumIp0?t=1788) Well, as we start to work with our community partners to try to figure out how to address some of those challenges, we think that the philanthropic opportunities need to be part of that mix. But a lot of that is going to be locally specific. So we hope to get to the place and we'd love to work with partners in other states. We've had an opportunity to demo this with some of our colleagues at the National Governors Association, some of my counterparts in other states, and we would love to partner with you to adapt this for your own needs. And I don't mean to speak for the Berne Center, but I know that they've been very generous in working with some of our counterparts across the country. Angie, do you want to add anything to that? So we are also planning to, as Quentin said,

**Anjith Prakash**

[30:35](https://youtu.be/IUbjzmumIp0?t=1835) we are also planning to share this out with other states. We are in talks with multiple states, but if there is any state, if you feel like you need this tool, you can always contact back to us

**Agueda Quiroga**

[30:46](https://youtu.be/IUbjzmumIp0?t=1846) and we can go through the process together. Okay, so there's a question by Paul. Have you seen any grant applicants move through your funnel, discover, analyze, draft, and submission?

**Anjith Prakash**

[31:03](https://youtu.be/IUbjzmumIp0?t=1863) So I can take that. So currently what we have done is, it's currently in the testing phase. Us as well as the FIO team has been using it a lot. We have been testing out. So currently we haven't submitted any grant applications, but we have been rigorously testing it out with current successful grants and whatever we could do with that.

**Agueda Quiroga**

[31:27](https://youtu.be/IUbjzmumIp0?t=1887) Okay, question by Laura. Have Rangwell apps been submitted as yet? If so, how does it compare to manually created apps by SMEs or consultants?

**Anjith Prakash**

[31:42](https://youtu.be/IUbjzmumIp0?t=1902) So as I said earlier, we haven't submitted applications yet, but then, so again, what I would say here is this is not a tool where we can write the whole grant and submit, because at the end of the day, the funds are a responsibility by a city or a state. So this promise can't be fulfilled by just an AI. There should be a human involved in it. So that's why we are trying to just help them in the starting phase, but then once they get into submitting or finishing it up, it will be all humans doing it.

**Agueda Quiroga**

[32:20](https://youtu.be/IUbjzmumIp0?t=1940) Great, thank you so much. We have another question by Josh. Are you aware of any restrictions within various grant programs against receiving AI-assisted materials?

**Anjith Prakash**

[32:34](https://youtu.be/IUbjzmumIp0?t=1954) So I think he's mentioning about the NIH regulations, which was put out. So what, if my assumption is correct, what they said was to not completely banning it, but restricting it, because some of the applications were having made up links or made up stories rather than like the actual story, which should come up in a grant. So that is something we are trying to avoid here by having this questionnaire. So I think this should be safer from it, but then always have a human in the loop, which would help to ease the process, I would say.

**Quentin Palfrey**

[33:16](https://youtu.be/IUbjzmumIp0?t=1996) Yeah, the humans have to be involved here. This has to be the quality control, the ownership, the ultimate, this mission has to be the product of their work. This is just helping them get to the starting point. And of course, we're going to have to stay in touch with our partners in the federal government agencies, make sure that we understand as they sort of think this through, as the space evolves, that we remain compliant with the needs of our federal government partners. But my sense is that this is a rapidly evolving field, and that we think that this is a good use case for AI to assist, but not replace human grant writers.

**Agueda Quiroga**

[33:58](https://youtu.be/IUbjzmumIp0?t=2038) Thank you so much. I have three questions in one by Joss that's very complex. So first one, which LLM do you use? How are you preventing private info getting outside your environment? Or are you hosting the LLM? And do you use RAG or train the model for specific capabilities?

**Quentin Palfrey**

[34:19](https://youtu.be/IUbjzmumIp0?t=2059) So I'm going to give all three of these to you, Angie.

**Anjith Prakash**

[34:23](https://youtu.be/IUbjzmumIp0?t=2063) Yeah, absolutely. And that's a great question, John. So the LLM, what we used as plot, and about the private information. So currently, we are using AWS as a cloud provider. And Massachusetts, Commonwealth of Massachusetts and AWS has a contract, where they are using a government cloud, where the data doesn't go out of the cloud. And we use Bedrock for our LLM calls, where all the cloud models are being hosted in Bedrock. So all the data goes to AWS, and it just comes back, it is never passed on to Anthropic or any companies. And this is a state contract. So if they send it out, it's going to be a huge issue. So that's where that goes through. And while talking about the RAG, yes, we are using RAG. And we didn't train a separate model, we are using RAG. And we are also planning to use some agents in the back end. So that would be the infrastructure. I hope

**Agueda Quiroga**

[35:21](https://youtu.be/IUbjzmumIp0?t=2121) I answered all the questions. I think so. Thank you so much. By Ronak, curious to know if this could have been used for the NOFO for the Rural Health and Transformation Program that all the states recently applied to?

**Anjith Prakash**

[35:44](https://youtu.be/IUbjzmumIp0?t=2144) So, yes, we could have used it, but I think it's already done. I'm not sure if it's closed yet. But then yes, we can definitely use this for the Rural Health Transformation Program.

**Agueda Quiroga**

[35:58](https://youtu.be/IUbjzmumIp0?t=2158) Another question about hallucinations. I mean, we all know about the hallucinations with output, but if we are uploading specific documents, are we restricting the hallucination output or not?

**Anjith Prakash**

[36:13](https://youtu.be/IUbjzmumIp0?t=2173) Yes. So RAG usually, AI usually hallucinates with output, as we all know. So uploading specific and we can also inject data into it. So how we are trying to get around it is by using RAGs and proper guardrails. So we have a dozen of PROMs and guardrails which try to put it inside. And that's how we are trying to not have hallucinations in our outputs.

**Quentin Palfrey**

[36:47](https://youtu.be/IUbjzmumIp0?t=2207) But it does underscore the importance of the human element in this process. We don't imagine this to replace the human applicants. The human applicants really need to do the quality control, need to read and review and really feel very comfortable with the truthfulness of the representations made in these grant applications. So there need to be human human reviewers and that needs to be a very robust process to avoid a situation where the AI makes a representation or a statement that isn't true.

**Agueda Quiroga**

[37:25](https://youtu.be/IUbjzmumIp0?t=2245) Thank you so much. Another complex question that has several parts. If my organization has a restriction on submitting to a public-hosted LLM, would your environment qualify as an exception to this? For example, not training on the data we submit and secure the other public

**Anjith Prakash**

[37:45](https://youtu.be/IUbjzmumIp0?t=2265) instances? Yes. So when we have a contract with AWS or whatever your cloud provider would be, they have their own system where they give a specific contract to the state saying that this data won't be used for training. So this, as I said, they will be hosted on their private virtual network. So you should be safe from any publicly exposed data or using that data for training their models.

**Agueda Quiroga**

[38:18](https://youtu.be/IUbjzmumIp0?t=2298) There are some questions into the chat that I am recovering, which are not in the Q&A. So I am going slowly because they are mixing with comments. So question by Sue, the tool taps into the federal and state resources to help the user. Where was the question again? Sorry. I mean, it's in the chat. It says the tool, I mean, well, taps into the federal and state resources to help the user. If you want to add more information to the question, feel free to unmute and ask again.

**Quentin Palfrey**

[39:00](https://youtu.be/IUbjzmumIp0?t=2340) So there's the capability to scrape, and Anjit can give a better technical explanation of how that works, but we're scraping the federal grant database in order to be able to do that sort of in a more automatic way. We would like to be able to do that with some of our own state grant databases. We have our own agency resources and we would like to be able to automate that. But there's also a mechanism for us to just manually upload the most common grant application, NOFOs, notices of funding opportunity, and put them into the tool and then have the tool be able to use them. And so in a sort of more manual way, we pulled in a number of the most frequent state grant opportunities and we're beginning to do some relevant charitable or philanthropic opportunities. Ideally, you get to the point where an applicant can use the search function and that would give you answers that would say there's this federal opportunity, but there's also...

[40:19](https://youtu.be/IUbjzmumIp0?t=2419) And that, you know, because a lot of the applicants that we deal with, they don't really care where the money comes from. And so being able to give them visibility into different sources and different opportunities can be quite helpful to them. That is especially true right now in a world where there have been so many disruptions to the federal program, but the scraping is easier or we're further along on the scraping with respect to the federal databases. I will say that as there have been disruptions in the federal funding environment, there also have been some disruptions in the information flow between agencies and between the agencies and the public. And so even the automated scraping is only as good as the information that is made public by the federal government agencies. Anjeet, do you want to add anything to the scraping or manual?

**Anjith Prakash**

[41:11](https://youtu.be/IUbjzmumIp0?t=2471) Oh yeah, absolutely. So currently how the scraping works is we are using the APIs from grants.gov. So we have developer access into it and they just give us these grant names along with the files, which is used for the NOFO and the deadlines and all those data. With that, we kind of get all the grants related to federal ones and we just ingest into our tool. And then we process

**Agueda Quiroga**

[41:38](https://youtu.be/IUbjzmumIp0?t=2498) it on tool and then it's displayed. Also in the chat, a very concise question. Can anyone use this tool and how?

**Anjith Prakash**

[41:51](https://youtu.be/IUbjzmumIp0?t=2511) So currently this is open to everyone in Massachusetts, but then as we move forward, we are planning to open it for other states, whichever states are willing to partner with us. We are working on those models, so hopefully it will be soon.

**Agueda Quiroga**

[42:09](https://youtu.be/IUbjzmumIp0?t=2529) Okay, so many questions into the chat. I am discovering now and not in the Q&A. Okay, I have another one in the Q&A, so I'm going first to the Q&A because it's where we are supposed to find the questions. So this is a question by Val. If more than one agency prepares application to the same grant program, is the AI system learning from or learning on both application responses such that application will start to have similar responses?

**Anjith Prakash**

[42:40](https://youtu.be/IUbjzmumIp0?t=2560) So no. Why? Because we have, so as I said, we have like 10 questions as the questionnaire. So these 10 questions are based on that specific city. So one could be city of Bradford, one could be city of Plymouth, for example. They both could have a different story to say. If it's high speed collision intersection or rebuilding the bike lanes, there could be different stories. So these questions are based on these stories, which is very specific to those cities. So when that comes out, the answers would be completely different and it would be similar.

**Agueda Quiroga**

[43:19](https://youtu.be/IUbjzmumIp0?t=2599) Okay, thank you so much. Another question from Alison. Is this tool offered in multiple languages?

**Anjith Prakash**

[43:28](https://youtu.be/IUbjzmumIp0?t=2608) So currently we only offer in English, but then if there is popular demand, we can always go for the languages, right Quinton?

**Agueda Quiroga**

[43:41](https://youtu.be/IUbjzmumIp0?t=2621) I'm trying to find the questions I have seen into the chat. Okay. So I think there's a question that has been repeated several times. So you have answered partially, but it's posted again. So it says, what safe words are in place to ensure AI-generated recommendations are accurate, defensible, and appropriate for public sector decision-making?

**Anjith Prakash**

[44:10](https://youtu.be/IUbjzmumIp0?t=2650) So as I said earlier, we have specified prompt for Massachusetts-related cities and everything so that that doesn't go past Massachusetts-related. And then we have guardian, so people don't misuse it for any other use cases. So that's how we kind of prevent it. And nowadays the AI models are also getting really well. So many of the guardrails just come from there. So, so far so good.

**Agueda Quiroga**

[44:36](https://youtu.be/IUbjzmumIp0?t=2676) Okay. So two more questions now to the Q&A. One by Sean. Are you capturing prior years scoring rubrics for the prior years opportunities for agencies and departments? Are there elements to help identify other related funding sources and help organizations build a capital stack and identity shortfall? Does this tool consider non-grant funding opportunities of tools such as long-term, as long-warranties, CDBG?

**Quentin Palfrey**

[45:17](https://youtu.be/IUbjzmumIp0?t=2717) I'm sorry, the last question was non-grant opportunities like a community development block grant or various kinds of formulaic resources. I think for the most part, this tool at this moment is focused on discretionary or competitive grant opportunities. And so, no, I think the answer is if there's a formulaic program like the community development block grant program or some of the monies that go from the U.S. Department of Transportation to each state transportation department kind of by formula and then are allocated out through some kind of a state-level decision-making process, I wouldn't imagine that that would appear in the tool as a resource because the decision-makers are different and that wouldn't be an opportunity that would come up. I think that we are trying to include non-federal sources of resources and that's a work in progress.

[46:20](https://youtu.be/IUbjzmumIp0?t=2780) As we mentioned, Anjith talked about scrapinggrants.gov. We're doing more manual processes for state opportunities and philanthropic opportunities, although there may be some scraping opportunities down the road for those. So does it cover the waterfront? No. Can it over time get a little bit better at identifying a universe of very helpful opportunities? Yes, I hope that we'll get a little bit better. Like any early stage tech tool, we're going to iterate and we're going to take feedback and try and get better over time at identifying more and more sources.

**Agueda Quiroga**

[47:04](https://youtu.be/IUbjzmumIp0?t=2824) Thank you so much, Quentin. There's a new question by John. Do you have an average inference or token cost input-output to the AI model charges for going through the process?

**Anjith Prakash**

[47:20](https://youtu.be/IUbjzmumIp0?t=2840) Currently, we are offering this free of cost for the users to use it and the government, the Massachusetts government is taking the cost and we are planning to do it like that. But to talk about the actual cost, it's the normal prices which the LLMs cost to the public. But yeah, the government is kind enough to give us access.

**Quentin Palfrey**

[47:50](https://youtu.be/IUbjzmumIp0?t=2870) We developed this tool in connection with a program that was focused on helping communities across Massachusetts and in particular communities that have been underserved historically overcome barriers to accessing federal funds. So we had some resources to work in this partnership to develop the tool. And the tool is being offered free of charge to our community partners. But our hope is that we can at least sort of let the infrastructure be available to others to adapt and evolve.

**Agueda Quiroga**

[48:23](https://youtu.be/IUbjzmumIp0?t=2903) Perfect. Thank you so much. There's a question in the chat from Sybil. I've had trouble with foundation funding not being captured by my grants management platform. Are there plans to expand the capture of NOFOs beyond grants.gov?

**Quentin Palfrey**

[48:43](https://youtu.be/IUbjzmumIp0?t=2923) So we touched on that a little bit. The answer is yes. We're absolutely trying to expand beyond the federal discretionary grant opportunities. We are a little bit further along in that with respect to state agencies in Massachusetts and what we've offered through our own agencies as well as our quasi-governmental agencies. We have begun to explore some of the philanthropic opportunities where we identify particularly attractive ones or widespread ones or ones that align with some of the programs that we hear the most about. We've started to incorporate those. But we'd be very open to hearing from folks if there are better ways of partnering with philanthropic organizations to get a little bit more coverage on that because we'd love to give more visibility into that. It's harder because it's a lot less sort of structured and systematic.

[49:44](https://youtu.be/IUbjzmumIp0?t=2984) And for a whole series of reasons, philanthropies will have very specific needs and desires in terms of what their grant catchment areas are or subject matter areas are. But we want to catch as much of that as we can. We think about this in a very user-centric way. And so we have very much in mind the community partners that we met on our roadshow as we sort of were asking folks, what do you need? And so if you think about it from their perspective, if you can find the philanthropic grants that meet with their needs and offer those to those community partners, that's really valuable.

**Agueda Quiroga**

[50:26](https://youtu.be/IUbjzmumIp0?t=3026) Thank you so much. I think we are close in time, but maybe there are a few questions to be answered. One is very short by Carol and it's in the chat and says, how will we know when many people can use it? So I think I did mention that.

**Anjith Prakash**

[50:47](https://youtu.be/IUbjzmumIp0?t=3047) So Quinton's team at FFIO is planning for a huge launch about Granville. So we are just waiting for that to come out, roll out. So there will be comms saying that, hey, Granville is out there. You can use it, play around with it. So just waiting for that.

**Quentin Palfrey**

[51:07](https://youtu.be/IUbjzmumIp0?t=3067) We will take some further steps to make this available outside of Massachusetts. There are some steps that we need to take before we do that. But we are very eager to work with folks to develop this. Our vision for this is absolutely for this to have an impact beyond Massachusetts and to be available to our partners. And so we'd love to work with you. We also would love feedback. We are making this up as we go along. And very proud of the work that we've done so far. But there's a lot of really dynamic people on this webinar. And we've learned a lot from you, even from your questions. And so we'd love to continue to partner with you and think about ways that we can take this project forward. I also want to correct something I said earlier. I must have had Boston University on the brain, working with them on something else. The Burn Center is at Northeastern and has been a terrific partner for us. So apologies for my misstatement earlier.

**Agueda Quiroga**

[52:08](https://youtu.be/IUbjzmumIp0?t=3128) No problem at all. So there's a lot of questions asking how to contact with the project. I mean, if you want to drop into the chat an email address or a way to contact, a lot of interest. And how do we provide this feedback and suggestions? This is another question. So I think that an email address or a contact permission. And maybe last question, totally, which is kind of technical again. It says, is it FOSS? I understand it only has opportunities for federal and Commonwealth of Massachusetts. But if we were to use it for fed grants and customize it ourselves with necessary license disclosure, of course.

**Anjith Prakash**

[52:56](https://youtu.be/IUbjzmumIp0?t=3176) So, yes, it is open source. It is available on GitHub. And yes, we are happy to work with them if they want to customize it for their own use case. Absolutely. We're open to it. I have dropped down my email ID in the chat. You can always reach out to me. We can have a talk and let's get it out there.

**Quentin Palfrey**

[53:16](https://youtu.be/IUbjzmumIp0?t=3196) I also would be remiss. So I want to thank Innovate US. I want to thank the Burn Center. I also want to thank some of the folks I see in the chat who were involved in setting this up. I see Jay, who was Anjit's partner in crime in developing this and going across the state. Our colleagues Mehar Jauhar and Mallory Strain are on the line. And our team has been very involved in this. I want to thank Beth Novick and the whole Burn Center team. It's been a real team effort. So I just want to give those shout-outs. But especially thank you to Jay, Mehar, and Mallory.

**Agueda Quiroga**

[53:50](https://youtu.be/IUbjzmumIp0?t=3230) Thank you so much for the session. I mean, it has been amazing. There's a lot of people asking if we are going to repeat the workshop once the tool is launched. So I think so. I mean, we are just scheduling it again. So thank you so much for everything. And thank you so much to our participants for all the interaction. I don't know, Anjit, Quentin, something final on the tool or something you'd like to share with our participants?

**Quentin Palfrey**

[54:16](https://youtu.be/IUbjzmumIp0?t=3256) No, thank you so much. But maybe our colleagues can put in the chat some ways to get in touch with both the Burn Center and our team.

**Agueda Quiroga**

[54:24](https://youtu.be/IUbjzmumIp0?t=3264) Thank you so much, Anjit.

**Anjith Prakash**

[54:26](https://youtu.be/IUbjzmumIp0?t=3266) Thank you so much. That's pretty much it. Thank you so much. I appreciate the support. And we'll be getting back to you.

**Agueda Quiroga**

[54:32](https://youtu.be/IUbjzmumIp0?t=3272) Thank you so much. And hope to see you soon. It has been a pleasure. I'm closing the session now. Bye.

* * *

Original page: https://innovate-us.org/introducing-grantwell-using-ai-for-municipal-technical-assistance · Licence: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org