I run a 24/7 autonomous agent on my own laptop, build multi-agent products end-to-end, and publish the failure modes I find along the way. If it needs an LLM, a queue, a webhook, and someone who will actually own the deploy, that's the work.
Live products, multi-agent hackathons, and the autonomous system I run myself. Every link below goes to a working artifact or a writeup of one.
A 24/7 autonomous agent running on Claude Code on my own laptop. FTS5-indexed memory across 90+ days of transcripts, a dual-judgment skill drafter that proposes new capabilities and blocks bad ones, a determinism-ladder architecture of 30+ hooks that enforce correct behavior at the shell level, and a Claude/Codex "boardroom" where two agents debate decisions before acting. It writes, commits, and deploys while I sleep.
Multi-agent medication reconciliation. 3 AI agents talk over the A2A protocol to cross-reference prescriptions against real FHIR health records, backed by 8 MCP tools for drug-interaction checks, dose validation, allergy screening, and FHIR output. Healthcare-grade inputs, auditable outputs.
AI research agent. Feed it a company name, it scrapes the web, synthesizes findings through Amazon Nova, and builds a semantically-searchable knowledge base. Two Nova models working together: one for synthesis, one for embeddings.
An honest diary that gets smarter the more you use it. Two surfaces, one product: a mobile journal for capture (voice + text, wikilinks, optional titles) and a web "Mirror" that reflects longitudinal patterns back at you. Voice transcription runs on Supabase Edge Functions + Gemini 2.5 Flash at ~$0.001/min.
AI agent that connects to your GitHub, Calendar, and Slack via Auth0 Token Vault, then gives you contextual briefings. Permission Control Center lets you grant and revoke access per provider.
Failed payment recovery for SaaS. Connect Stripe, get smart dunning emails, expiring card alerts, and recovery analytics. Multi-tenant with Stripe Connect.
Customer feedback boards with upvoting, public roadmap kanban, changelog with email notifications, and an embeddable widget. A Canny alternative at a tenth of the price.
Real-time AI interview coach. Listens to your interview via Gemini Live API, analyzes body language with Vision, and gives live coaching feedback with scoring.
Free, open-source EMDR bilateral stimulation tool. Visual, audio, and haptic modes with sine-wave motion, set timer with breathing rest intervals, and full customization. No account, no tracking, no paywall.
AI form navigator using Gemini Vision. Analyzes any web form and overlays intelligent guidance on each field. Understands context, suggests values, flags required fields.
The tools depend on the problem. These are the ones I reach for most.
TypeScript, Python, SQL. Production code in all three, daily.
Next.js, React, Tailwind, shadcn/ui. Server components by default.
FastAPI, Node.js, PostgreSQL, Supabase. REST and webhooks.
Claude, GPT, Gemini APIs. RAG pipelines, embeddings, tool calling, agents.
AWS (EC2, Bedrock, Lambda), Vercel, Docker, Cloudflare. CI/CD.
Stripe, Stripe Connect, webhooks, subscription billing, invoicing.
Technical writeups of systems I actually shipped, and post-mortems on what broke.
Tell me what you need. I'll reply within 24 hours with a straight answer on whether I can help and what it would take.
or email [email protected]