The models will keep getting better. What matters is the harness — the infrastructure that makes them useful. Think of it like this: a language model is a brain, but a brain without hands, eyes, or a voice can’t do anything for you. The harness is the agents, voice pipelines, memory systems, and APIs that give AI the ability to actually operate inside your business. We’re model-agnostic — the best model for the job gets the job. Everything below is running in production right now, except Hard-E, which is demo on request.
My CRM for home services and exterior contracting, MVP live with a launch customer. Property research and photo or Street View analysis run on each new inquiry.
Voice-first AI agent. 21 CRM tools, 3-tier memory, real-time WebRTC voice. Multi-tenant, cloneable. Demo on request.
Technical write-up →Care coordination. Circles, tasks, digests, helper share pages. MCP server and REST API, so you can run a circle through your own agent. Free, bilingual, live.
See it live →AI recovery companion. Voice-enabled, own memory, reads/writes real recovery data. 10 modalities.
See it live →My always-on agent, first on OpenClaw and now on Hermes Agent. Takes voice or text requests over Telegram, writes code with Claude Code, publishes sites, keeps research wikis, helps coordinate care through the OpenCare API, and watches its own costs.
Serverless content pipeline. AI colorization, captions, multi-platform posting. Zero human touch.
See it live →Python, JavaScript/TypeScript, C#, SQL, Bash
Next.js 14, React, FastAPI, Node.js, Tailwind, Unity, Telegram Bot API, boto3
Model-agnostic. Claude (Sonnet, Haiku, Opus, Code), GPT-4o, GPT Vision, xAI Grok, Perplexity, Deepgram, Cartesia, ElevenLabs, Pipecat
EC2, Lightsail, Lambda, S3, DynamoDB, SES, CloudWatch, Amplify
Docker, Nginx, PM2, systemd, SSL, Fail2Ban, zero-downtime deploys
Supabase (PostgreSQL + RLS), DynamoDB, Redis, Clerk, Persona, Nostr