Projects

Systems work with a visible point of view.

A mix of professional and personal work across agentic AI, backend platforms, orchestration, and robotics. I care less about novelty theater and more about whether the system survives real use.

Talk through a project
Professional / Agentic AI

Recipe support agent · Intuit

Designed and built an agentic Slack support bot for Recipe, Intuit's data curation and developer platform. It answers from a RAG knowledge base first, then investigates with tools when evidence is missing.

  • LangChain orchestration loop with Claude via Intuit GenOS and SageMaker
  • Custom MCP server exposing internal diagnostics and support actions as tools
  • RAG over code, docs, Slack history, meeting transcripts, and runbooks
  • Escalation path only when investigation cannot confidently resolve the issue
PythonLangChainClaudeMCPRAGSlack
Personal / Agent platform

Clotho

Greenfield multi-tenant agent platform built around compiled-plan orchestration: an LLM plans a typed execution graph and a deterministic interpreter runs it with human approval before execution.

  • pgvector memory with secret-redaction before embedding and per-tenant retrieval
  • BYOK AES-256-GCM credential vault and multi-provider LLM routing
  • Durable execution and human-in-the-loop control via Hatchet
  • 21 packages with TDD-first workflow and architecture decision records
GopgvectorHatchetAES-GCMPostgres
Personal / Creative AI system

chotu.studio / auto-shorts

AI studio that turns a creative brief into a finished short through a seven-stage orchestrated DAG covering story, characters, cinematography, frames, motion, audio, and assembly.

  • Durable run-ledger for crash recovery without double-billing
  • Multi-provider generative-media integration behind a unified adapter
  • Per-run cost and latency metering
  • 545 backend tests and 458 frontend tests documented in the project README
PythonFastAPIReactMCPClaude
Professional / Robotics platform

Genesis / Gaia / ApexDrive

At MachaniRobotics I worked across the software spine of humanoid robotics: gateway services, protocol orchestration, facial-recognition pipelines, and a C++ limb-driver controlling motion in the hardware stack.

  • Go gateway with gRPC, GraphQL, protobuf, Nginx, and Envoy
  • C++ limb-driver for precise humanoid movement control
  • Facial-recognition pipeline improvements through embeddings and augmentation
  • Conversational robot interface using ChatGPT APIs with TTS and STT
GoC++gRPCGraphQLEnvoyNginx