Projects

NeatCode

A code graph that answers coding agents' structural questions over MCP instead of making them read whole files.

to index 131k LOC

3.8s

structural queries

<300ms

incremental re-index

0.49s

TypeScript · ts-morph · SQLite · MCP

Distributed speech-to-image pipeline

Live narration in, illustrated story panels out, fanned out across a fleet of GPU pods.

scene candidates rendered in parallel

3

GPU instances in production

4

tests

115

Python · FastAPI · Celery · Redis · PostgreSQL · MinIO · ComfyUI · WebSockets · AWS EC2 GPUs

Fleet

GitHub and Booking.com rebuilt as environments for agent evaluation and RL.

RL environments · Agent evaluation

Generative AI platform

Model routing, tool calling, a custom MCP server, and RAG over vector databases.

response relevance (RAG)

+25%

LangChain · MCP · Vector databases · RAG

Ivorie

Backend for a three-sided wedding marketplace: real-time chat, milestone payments, vendor payouts.

users

5k+

messages a day

10k+

NestJS · Socket.IO · Redis · PostgreSQL · Prisma · Stripe Connect · Expo React Native · Next.js · AWS

Multi-tenant AI app platform

One Go + TypeScript backend serving several consumer AI apps, isolated per tenant, shipped with blue-green deploys.

apps on one backend

2

LLM providers

3

commits, ~80% mine

289

Go · Gin · GORM · PostgreSQL · pgvector · Redis · TypeScript · LangGraph · OpenTelemetry · GitHub Actions

mcp-evals

An open-source harness being built to benchmark MCP servers across models: accuracy, tool choice and context cost.

graded runs in the first real benchmark

144

less context with NeatCode (GPT-6.1 Sol)

61%

lower cost per task, same comparison

46%

TypeScript · MCP SDK · zod · Commander