Skip to content
DB, Dhruv Kamalesh Kumar, back to top

00 / INTRO

00 / INTRO

Dhruv Kamalesh Kumar, AI Engineer

AI Engineer

The Burnes Center for Social ChangeNortheastern University

Boston, from Bangalore

Shipping production software since 2021

I ship AI products end to end, from first prototype to thousands of real users.

  • 375+IEPs read by A‑IEP
  • 16,148Monthly active users on Acharya ERP
  • 2ndPlace, AWS x Riot Games hackathon
See the work

Take the 45-second tour

01/Work

08 selected

Selected work

Production systems, one hackathon podium, a tool on npm and the app where it started. Each one opens into a full case study with the problem, the architecture and the receipts.

  1. 01

    A-IEP

    Director's pick2025 to nowGovernment AI
    Ownership: Led the engineering

    A special‑education plan, in words a parent actually uses.

    4Languages in production

    I ownedLead engineer who took the earlier co-op prototype to production: I built the processing pipeline and the privacy design.

    LiveA special‑education plan, in words a parent actually uses. 4 Languages in production. 2025 to now, Government AI. Led the engineering. I owned: Lead engineer who took the earlier co-op prototype to production: I built the processing pipeline and the privacy design.
  2. 02

    GENIE

    Director's pick2024Government AI
    Ownership: Core contributor on a team build

    A safe place for state employees to try generative AI.

    44,000+State employees with access

    I ownedCore engineer on the Burnes team’s build, and the author of Smart Model, the router that picks a model for each request.

    A safe place for state employees to try generative AI. 44,000+ State employees with access. 2024, Government AI. Core contributor on a team build. I owned: Core engineer on the Burnes team’s build, and the author of Smart Model, the router that picks a model for each request.
  3. 03

    arc-control-mcp

    Director's pick2026Tooling
    Ownership: Sole author

    An MCP server that lets agents drive the Arc browser.

    50Tools

    I ownedSole creator: all 50 tools, the rule that an agent never touches the tab or window I am using, and the CI and npm release.

    An MCP server that lets agents drive the Arc browser. 50 Tools. 2026, Tooling. Sole author. I owned: Sole creator: all 50 tools, the rule that an agent never touches the tab or window I am using, and the CI and npm release.
  4. 04

    Civic AI Course Delivery

    Director's pick2026Platform
    Ownership: Sole author

    An 8‑day, SMS‑paced course platform for InnovateUS Civic AI.

    3Days, solo, to a working course

    I ownedSole engineer, start to finish in three days: sign‑in, SMS pacing, the quiz gate, certificate, admin console, runbook and the carrier compliance doc.

    An 8‑day, SMS‑paced course platform for InnovateUS Civic AI. 3 Days, solo, to a working course. 2026, Platform. Sole author. I owned: Sole engineer, start to finish in three days: sign‑in, SMS pacing, the quiz gate, certificate, admin console, runbook and the carrier compliance doc.
  5. 05

    ABE and One-L

    Director's pick2025 to 2026Government AI
    Ownership: Led the engineering

    Two AI agents for state procurement, recognized by NASPO in 2025.

    2NASPO awards for ABE and One‑L, 2025

    I ownedJoined One‑L in November 2025 and ABE in February 2026, after the 2025 NASPO awards. Since then I lead ABE’s engineering, where I built the quality monitoring and the thumbs-up-to-test-case pipeline; on One‑L I set the direction, with four co-op students on the build.

    Two AI agents for state procurement, recognized by NASPO in 2025. 2 NASPO awards for ABE and One‑L, 2025. 2025 to 2026, Government AI. Led the engineering. I owned: Joined One‑L in November 2025 and ABE in February 2026, after the 2025 NASPO awards. Since then I lead ABE’s engineering, where I built the quality monitoring and the thumbs-up-to-test-case pipeline; on One‑L I set the direction, with four co-op students on the build.
  6. 06

    VCT Scout

    Director's pick2024Hackathon
    Ownership: Core contributor on a team build

    Ask in plain English, get a Valorant roster. Second place worldwide.

    2ndPlace overall

    I ownedOne of four on a hackathon team. I designed the agent’s tool calling and wrote the system prompt that orchestrates it, plus the tools that save and load a team composition; teammates built the chat backend, the player-data tools and the map tools.

    Ask in plain English, get a Valorant roster. Second place worldwide. 2nd Place overall. 2024, Hackathon. Core contributor on a team build. I owned: One of four on a hackathon team. I designed the agent’s tool calling and wrote the system prompt that orchestrates it, plus the tools that save and load a team composition; teammates built the chat backend, the player-data tools and the map tools.
  7. 07

    Public Voice

    Director's pick2026Government AI
    Ownership: Led the engineering

    A voice survey that asks one smart follow‑up, then lets you go.

    LiveIn production at InnovateUS

    I ownedTechnical lead: I set the technical direction and built the 2025 voice survey prototype it grew from. Two teammates built most of the production code.

    LiveA voice survey that asks one smart follow‑up, then lets you go. Live In production at InnovateUS. 2026, Government AI. Led the engineering. I owned: Technical lead: I set the technical direction and built the 2025 voice survey prototype it grew from. Two teammates built most of the production code.
  8. 08

    Acharya ERP

    Director's pick2021 to 2023Mobile
    Ownership: Sole author

    Where the shipping started: one Flutter app, thousands of students.

    5.4×Monthly active users while I owned it

    I ownedThe Flutter app, start to finish, then the handoff to the interns I trained.

    Where the shipping started: one Flutter app, thousands of students. 5.4× Monthly active users while I owned it. 2021 to 2023, Mobile. Sole author. I owned: The Flutter app, start to finish, then the handoff to the interns I trained.

The lab

Everything else

Shared infrastructure, tooling, mobile work and experiments. Smaller in scope, same habits.

02/Impact

Numbers, and who says so.

Each figure says who stands behind it: a third party, my employer, or me. Where there is a public page, it links.

All receipts

IEPs read by A-IEP

375+

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.

From the reposource: github.com

  • Monthly active users on Acharya ERP

    16,148

    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.

    Self-reportedsource: play console

  • State employees with access

    44,000+

    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.

    Employer-reportedsource: burnes.northeastern.edu

  • AI tools shipped

    26

    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.

    Employer-reportedsource: burnes.northeastern.edu

  • Engineers mentored

    50+

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

    Self-reportedsource: resume

  • Place, AWS x Riot Games hackathon

    2nd

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

    Third partysource: press.aboutamazon.com

  • 2

    NASPO awards for ABE and One-L

    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.

    Third partysource: naspo.org

03/Proof

Proof, in other people’s words

5 awards8 press features4 talks

Awards

As covered by

Press and talks

04/How I build

Built so people can rely on it

Pipeline: A-IEPState machine

The path every IEP takes: names redacted before a writing model reads a word, the original deleted, and a purge on any failure.

0102030405

  1. Upload

    S3, Step Functions

    A parent’s IEP lands in an encrypted bucket and a Step Functions run starts.

  2. OCR

    Mistral OCR

    The one step that sees the original page, names included. Mistral’s copy is deleted once it answers.

  3. Redact

    Comprehend PII redaction

    Every identifier except dates is replaced, and an error stops the run. Then the original upload is deleted.

  4. Summarize

    OpenAI GPT-4.1

    Reads redacted text only and writes the summary in nine sections, each pointing back to its pages.

  5. Translate

    OpenAI GPT-4.1

    Spanish, Vietnamese and Chinese, from the English summary. Any failure purges unredacted artifacts.

Principles

  1. 01

    Software is finished when someone who is not an engineer can rely on it.

  2. 02

    Privacy is a design input. I redact names and other identifiers before the analysis model reads a word.

  3. 03

    I measure the model before I trust it, and I say so when I cannot yet.

What I reach for

Models and agents
  • Claude via Amazon Bedrock
  • OpenAI
  • Mistral OCR
  • RAGAS evaluation
Cloud
  • AWS CDK
  • Step Functions
  • Lambda
  • Comprehend
Product
  • TypeScript
  • Python
  • React
  • Next.js

Built with agents

Building with agents since February 2025. Here is the log.

Activity strip, 28 Feb 2025 to 20 Oct 2026: 235 active days out of 600. The busiest day was 16 Apr 2026, with 3,484 model turns. Each day is shaded by how many turns the agents took.

Tool calls
55,660in 113,952 model turns
Sessions
808plus 563 subagents
Active days
235of 600 in the window
Output tokens
22.8MClaude Code, from 2,610 prompts

This site and arc-control-mcp, an MCP server on npm that lets agents drive the Arc browser, were built this way.

Self-reportedsource: local Claude Code, Cursor and Codex logs28 Feb 2025 to 20 Oct 2026

From what this Mac kept: 655 Cursor chats, 139 Claude Code sessions and 14 Codex sessions. Across the three, roughly 13B to 28B tokens went through the models, nearly all of it cached context read back in on each turn. That is why output tokens, Claude Code's own count, are the figure above.

05/Director

Don’t scroll.
Direct.

Pick a guided tour. The site scrolls, spotlights the work and redraws the particles behind this page.

Choose a guided tour

Enter to begin / Esc to stop

A tool-calling model plans each tour live, and speaks lines I recorded.

06/About

Bangalore to Boston

From shipping a Flutter app to thousands of students to building AI for government.

Bangalore12.97°N 77.59°EBoston42.36°N 71.06°W

  1. 2018

    Bangalore, India

    B.E. in Computer Science

    Acharya Institute of Technology

    Four years of computer science, and a lot of Flutter on the side. I graduated in July 2022.

  2. 2021

    Bangalore, India

    Software Engineer, Flutter

    Acharya Institutes

    Built the Acharya ERP app in Flutter. Every student and staff member used it; monthly active users grew from 2,972 to 16,148 while I owned it, and I trained interns to take it over.

  3. 2022

    Boston, MA

    Crossing oceans

    Northeastern University

    Moved to Boston for an M.S. in Artificial Intelligence at Khoury College. Computer vision, NLP and deep learning, with a 3.83 GPA.

  4. 2024

    Boston, MA

    GenAI Product Development Co-op

    The Burnes Center for Social Change

    Joined the AI for Impact co-op in January. Built the Generative AI Sandbox for Massachusetts state employees with EOTSS and AWS, and wrote Smart Model, which picks the right model for each request.

  5. 2025

    Boston, MA

    Lead AI Engineer on A-IEP

    The Burnes Center for Social Change

    Took the AI‑EP prototype to production as A‑IEP, an open source platform that helps parents read their child’s IEP in their own language. I made my first commit in February.

  6. 2026

    Boston, MA

    From shipping to measuring

    The Burnes Center for Social Change

    Public Voice and the Public Engagement Coach for InnovateUS. In July I audited A‑IEP, found no automated check on accuracy, and started a synthetic IEP benchmark.

07/Off duty

Off the clock, still aiming.

Valorant and CS2 are how I reset. I used to stream it, and I still aim fine.

Games
Valorant and CS2
From
Bangalore, India
Past life
Streamed Valorant and PC builds on YouTube. 181 videos, about 1,270 subscribers, sponsored by Epic Games, GeForce NOW and Colorful
Desk
Builds PCs and mechanical keyboards, always retuning the setup
Reading
One Piece. A whole wall of manga panels, one day at a time
Eats

Workshop
3D printing whatever the desk needs next

Currently building

  • 01A synthetic IEP benchmark, so A‑IEP can report measured accuracy
  • 02Course Delivery SMS: an 8‑day, text‑paced video course for InnovateUS
  • 03Public Voice, a no‑login voice survey that asks one smart follow‑up
  • 04arc-control-mcp, a 50-tool MCP server that lets agents drive the Arc browser

08/Contact

Building something early? Let’s talk.

I work fast, in small teams, and I ship things people use.

Tell me what you are building, what stage it is at, and what you need help with first.

Write to me

© 2026 Dhruv Kamalesh Kumar

Boston --:-- ---

Built with Next.js, three.js, GSAP. Designed and engineered by DB.