Every number, with its receipt

Plans processed in production since launch, from about 260 family accounts (July 2026). A parent who can read the plan can speak up in the meeting. I lead A‑IEP’s engineering.

• Home, Impact
• Home, Hero
• A-IEP, outcomes

Android, from the Play Console: 2,972 when I took it over on 1 Sep 2021, 16,148 when I handed it over on 31 Aug 2023. Every student and staff member used it. Where I learned that when an app breaks, a real person has a bad day.

• Home, Impact
• Home, Hero
• Home, About

A sanctioned place to try AI is where a government starts using it. This counts access, not daily use. I co-built the sandbox behind the number.

• Home, Impact
• GENIE, outcomes
• Home, Work

A program total, not mine alone: the AI for Impact program has built 26 tools for government and civic partners. I help lead its technical side.

• Home, Impact

Student engineers taken from prototype to production deployments across the program. A headcount across the program, not a measured outcome.

• Home, Impact

VCT Scout also won Best Cross-Regional Team Submission, built in a hackathon sprint with three teammates.

• Home, Impact
• Home, Hero
• VCT Scout, outcomes
• Home, Work

The Cronin Gold Award and the Academic Collaboration Award, won in 2025 by Massachusetts OSD with the Burnes Center, before I joined either project. I now lead ABE's quality monitoring and One-L's direction.

• Home, Impact
• ABE and One-L, outcomes
• Home, Work

English, Spanish, Vietnamese and Chinese are enabled in production. Arabic, with a right-to-left layout, is built and enabled on dev and staging only.

• A-IEP, outcomes
• Home, Work

Parents co-designed the tool and reported feeling more confident advocating for their children. Reported by the Burnes Center and Reboot Democracy, October 2026.

• A-IEP, outcomes

69 TypeScript and 31 Python test files in the public repo, including the pipeline’s failure-path tests and a Playwright end-to-end suite.

• A-IEP, outcomes

One router call reads the prompt against a capability sheet of task fit, cost and token size, and picks the model. Simple requests can go to models up to 12 times cheaper per request than the default: Claude 3 Haiku against Claude 3.5 Sonnet at Bedrock list prices, for 1,500 tokens in and 500 out. The list is in the router's own code.

• GENIE, outcomes

Signed by Governor Healey and Lieutenant Governor Driscoll on June 26, 2024.

• GENIE, outcomes

In twelve modules, as of version 0.4.0.

• arc-control-mcp, outcomes
• Home, Work

Cover the tool schemas, argument validation, the agent window and focus guard, the DevTools engine against a real Chromium, and live runs against Arc. There is no benchmark of agent task completion.

• arc-control-mcp, outcomes

Month ending October 1, 2026. Registry counts include automated installs.

• arc-control-mcp, outcomes

From the first import on 29 Sep to the admin console on 1 Oct 2026, one engineer end to end.

• Civic AI Course Delivery, outcomes
• Home, Work

The real chapters of the InnovateUS Civic AI course, about 34 minutes of video in all. Answers are checked on the server; the answer key is pending author sign-off before launch.

• Civic AI Course Delivery, outcomes

Not public yet, and the repo records no launch date. Dev and prod stacks, alarms, a CI deploy and a runbook are in place. Still open before launch: re-file the carrier campaign with the current wording, confirm the monthly spend limit and get the answer key signed off.

• Civic AI Course Delivery, outcomes

From my résumé: lawyers review a redline that One‑L has already drafted, instead of marking up each contract by hand. The baseline and sample are not published here.

• ABE and One-L, outcomes

A separate award the team also won.

• VCT Scout, outcomes

The size of the data the team restructured so Athena could answer in time, not a measure of accuracy.

• VCT Scout, outcomes

Open at publicvoice.innovate-us.org, no login needed.

• Public Voice, outcomes
• Home, Work

Answers are transcribed and personal details are redacted before anything is saved.

• Public Voice, outcomes

The survey is built to take about three minutes.

• Public Voice, outcomes

Android, from the Play Console: 2,972 monthly active users on 1 Sep 2021, when I took it over, and 16,148 on 31 Aug 2023, when I handed it over. Every student and staff member across Acharya Institutes used it.

• Acharya ERP, outcomes
• Home, Work

Android, 21 Aug 2023. Daily active users averaged about 370 in my first month and about 6,900 in my last.

• Acharya ERP, outcomes

From 1.2 stars to 4.5 while I owned the app.

• Acharya ERP, outcomes

The GovLab nine-question framework the coach is built around.

• Public Engagement Coach, outcomes

At 103 ms latency.

• Home, About

The first stage I measured is OCR, on synthetic IEPs (clean, scanned and photographed), scoring digit-bearing facts, checkboxes and invented numbers rather than text similarity. A self-hostable 1B model, LightOnOCR-2, matched production Mistral OCR on facts, numbers and checkboxes and invented no numbers; the only difference was one dropped diacritic in an 8 pt table. A 4-bit build of olmOCR 2 read 0% of checkboxes on clean pages while its text similarity stayed near 97%; the FP8 build read 100%. Quantization, not the model, removed the information parents act on. Runs from 30 Sep and 1 Oct 2026; the results are not published yet.

• A-IEP, production notes

For the summaries themselves, the only automated check is still a schema check: nine sections present, in the right shape. The summary benchmark is a written design, not a result. It scores atomic facts against both the summary and the source, uses a judge from a different model family, and grades each translation with a per-language judge. No accuracy number exists yet, so none is shown.

• A-IEP, production notes

Judge-based scoring is projected at about $0.25 to $0.55 per document, or $25 to $110 a month at 100 to 200 documents. Those are planning estimates from my cost study, not a bill. A comment in the pipeline’s tests records the median of 100 production and staging runs: 29 seconds to summarize and 24 to translate.

• A-IEP, production notes

I never measured a saving, so none is claimed. Picking a cheaper model for simple prompts saves money, but the router call itself runs on Claude Sonnet and reads the whole capability sheet, so on a typical request it costs about as much as it saves. A small, cheap router model is the first fix I would make.

• GENIE, production notes

The repo has no tests and no routing evaluation, so nothing shows that routing kept answers as good as always using one model.

• GENIE, production notes

The Burnes Center repo is private, so the router is described here and not linked.

• GENIE, production notes

Sixteen test files cover the advertised tool schemas, argument validation, timeouts, batch limits and an integration run. There is no benchmark of how often an agent completes a task with it.

• arc-control-mcp, production notes

29 server test files cover the reminder rules, the watch-coverage check, text wording and length, STOP handling, sign-in limits and the admin API. The course is pre-launch, so there are no usage numbers.

• Civic AI Course Delivery, production notes

The repo is private, so nothing here is linked. Status: pre-launch.

• Civic AI Course Delivery, production notes

A Step Functions pipeline scores each test question on seven RAGAS metrics, including faithfulness and context recall. Test questions are curated, plus thumbs-up answers that an LLM rewrites into Q&A. The judge is the same primary Claude model that writes the answers, so the scores are not cross-family. No results are published here, and the screenshot below is sample data.

• ABE and One-L, production notes

A hackathon build: no test set, and no evaluation of whether a quoted stat matches the data.

• VCT Scout, production notes

The production repo is private. The 2025 prototype it grew from is public.

• Public Voice, production notes

No evaluation of the routing or the Pro Tip judge is in the repo.

• Public Engagement Coach, production notes

The repo is private. I was technical lead; teammates wrote most of the code.

• Public Engagement Coach, production notes

Commonwealth of Massachusetts, signed by Gov. Maura Healey and Lt. Gov. Kimberley Driscoll. For developing and iterating on AI-based solutions to advance state services. Dated June 26, 2024.

• Home, Proof

NASPO. Won by Massachusetts OSD for its dual AI agents, ABE and One‑L, built with the Burnes Center. The award came before I joined: I started on One‑L in November 2025 and ABE in February 2026, and now lead ABE’s quality monitoring and One‑L’s direction.

• Home, Proof

NASPO. For the OSD and Northeastern partnership behind the procurement AI tools. Also before I joined the work.

• Home, Proof

AWS and Riot Games. VCT Scout, with Aravind Dasarathy, Rudra Sett and Akshay Gunjur Surya Prakash. More than 3,200 participants.

• Home, Proof

AWS and Riot Games

• Home, Proof

Amazon Press Center

• Home, Proof

Boston Business Journal

• Home, Proof

Northeastern Global News

• Home, Proof

GovTech. Covers the InnovateMA projects, including the Generative AI Sandbox. Quotes me as Dhruv Kumar.

• Home, Proof

Massachusetts Governor’s Office

• Home, Proof

Mass.gov

• Home, Proof

Reboot Democracy. My post (February 3, 2026) on how parents saw the A‑IEP prompts and reshaped them.

• Home, Proof

InnovateMA, Massachusetts State House. The AI for Impact co-op cohort showed its work to the Governor in June 2024.

• Home, Proof

InnovateUS Prompting Lab. I taught this 45 minute workshop on March 27, 2026. Participants prompted AI step by step to build a working website with no coding background.

• Home, Proof

InnovateUS Prompting Lab. Live Q&A on April 10, 2026, following the coding workshop. InnovateUS lists me as a speaker.

• Home, Proof

Reboot Democracy. Published October 1, 2026 on the A‑IEP pilot in San Francisco, where more than 200 families took part.

• Home, Proof

AWS, Boston Seaport. I walked through the MassHealth Helper design using the Well-Architected principles at the AWS Boston Seaport office on September 6, 2024, alongside the AWS account team.

• Home, Proof

I moved to Boston for an M.S. in Artificial Intelligence at Northeastern (GPA 3.83, December 2024).

• Home, About

I joined as a co‑op in January 2024 to build the Generative AI Sandbox for Massachusetts state employees, and I have been full-time since July 2024.

• Home, About

Streamed Valorant and PC builds on YouTube. 181 videos, about 1,270 subscribers, sponsored by Epic Games, GeForce NOW and Colorful

• Home, Off duty

More than 3,200 participants.

• Home, Proof