CHIRAG AGARWAL

I make AI reliable, and easier to build.

Senior AI platform engineer and tech lead for Hinge Health's care agents.

At work
Turning one-off AI agents into a platform: shared building blocks other teams build on, evals that catch what matters, and a clinical-safety and conversation-quality bar every agent has to clear.
At home
Turning gaps in my own day into apps I use daily: built with a small team of coding agents I lead, each one a lab for what AI can do next.
Scroll · Notes · Projects · Work ↓
01

Notes

Short write-ups of design decisions after they ship: the problem, the options I tried, what I picked, and who or what shaped it.

Projects

Apps I build with agents, for myself, every day. The code is private; each has a page on how it was designed and why.

Shruti

Ragas for harmonium and Mohan veena

A raga companion for harmonium and Mohan veena, built for playing by ear: see a raga, hear it, and find out what you're playing. Fully offline.

Shruti app icon SHRUTI · 2023
2023 · Expo· TypeScript· Canvas Read the case study

Bloom

A keyboard for thumbs

A hex keyboard for Android, designed for thumbs, with Learn to make the switch from QWERTY. No internet access.

Bloom app icon BLOOM · 2024
2024 · Kotlin· Canvas· Android IME App page Read the case study

Mira

A language partner I'm building

A language trainer for Android: kanji drawn stroke by stroke on demand, and a spoken-French partner in the works. Building it is half the point.

Mira app icon MIRA · 2026
2026 · IN PROGRESS · Agents· Expo· Vercel· LangSmith· ElevenLabs Read the case study

Earlier

Experiments and tools from before, kept for the record.

2025

Oracle A self-hosted model gateway

REST access to cloud and local models, with agents on top and GGUF models running locally on KoboldCPP.

2025

EchoOS A voice-first operating system

A concept OS for small, screenless, always-on devices: voice, memory and context instead of a screen. Offline and private.

2025

Agent framework Multi-context agents

Extensible agents that keep context across tasks, with a scheduler, on Ollama or cloud models.

2025

Creativity Vocal training, Nusrat-inspired

A concept for AI vocal training and remixing, inspired by Nusrat Fateh Ali Khan: pitch tracking, style transfer, a 2D pitch map.

2025

Weights visualizer Looking inside transformers

2D and 3D views of transformer weights and biases, with UMAP to find structure.

2026

ebook-tools Reading on e-ink

PDF to EPUB with e-ink covers, EPUB repair, sleep screens and a Wi-Fi client for an Xteink X3 e-reader.

03

Work

Senior AI Platform Engineer, Tech Lead · Hinge Health

I work on the AI platform at Hinge Health: the agents, the evals, and the building blocks other teams build on. Before that, seven years of event-driven backends and platforms.

NOW
DEC 2024 – NOW

Senior AI Platform Engineer, Tech Lead · Hinge Health

On the AI platform team behind Robin, Hinge Health's AI care assistant.

  • Led the move from a monolithic AI agent to a modular LangGraph system, and broke a monolithic context builder into independent, testable data modules.
  • Lead the next-generation exercise recommendation engine as PM and engineer. It is one shared capability any agent can call, it moved from rules to conversational, personalized recommendations, and it asks before it changes anything.
  • Made clinical alignment part of shipping: a sign-off loop with clinical partners, claims grounded in vetted content, fabrication paths closed, and judges that track clinical accuracy and safety over time.
  • Designed the assistant's evaluation suite on LangSmith: the judges, the dataset flow, and one lean whole-conversation trace that every agent emits.
  • Architected RAG and vector search infrastructure on Databricks.
  • Built multimodal LLM features for text and voice, with HIPAA-compliant integrations.
  • Own release engineering for a multi-team AI service, from release branches to continuous deployment. I run AI coding agents as a supervised team, so every PR is agent-reviewed and verified before a person sees it.
NEXT

Platformization

  • Turning the agent building blocks into packages other teams can build on: one place to choose an LLM provider, and a base agent with tracing built in.
  • Agents that can't be pointed at the wrong person's data. Identity comes from a verified token, never from a parameter the model fills in.
  • Evals that say what to fix, not just that something broke.
BEFORE
APR 2022 – DEC 2024

Senior Software Engineer · Hinge Health

  • Event-driven NestJS services and Kafka; led a migration to Databricks that saved $2M a year.
2019 – 2022

Software Engineer II · Navis

  • Real-time yard health dashboard, Node microservices, WebSockets.
2018 – 2019

Graduate Research Assistant · San Francisco State University

  • Security analysis of hybrid mobile apps across 2M apps. MS in Computer Science.
AT WORK AND AT HOME
Shared building blocks other teams build onShared skills every project's agents work from: spec, plan, build, ship
Release engineering with agent-reviewed PRsGoogle Play releases run by a release agent, from signing to privacy pages
Parallel execution across teamsParallel agent sessions, with me as the lead

The end · keep scrolling to start again