Projects

A group cannot agree where to meet.

Picks a meeting point that is fair to everyone and explains why each venue ranked where it did. It measures fairness in real commute time instead of straight-line distance.

FastAPI / React / Maps APIs

Every factory defect has to end up as ISO 9001 paperwork.

Three agents turn a defect image into CAPA and PFMEA documents that follow ISO 9001, with a human supervising the output.

Multi-Agent / Vision LLMs / ISO 9001

A rough product idea is not something you can build from.

Turns the idea into a PRD, architecture diagrams, and AI coding prompts in a few minutes.

Gradio / Qwen / Prompt Engineering

Bill photos hold financial data nobody types in.

Reads bill images with a vision LLM and turns them into spending insights.

Vision LLMs / Python

Making a short video takes six different jobs.

Six agents split scripting, storyboard, voice, visuals, editing and review, and return a finished video from one prompt.

Multi-Agent / Video Generation

When an AI agent fails, nobody can prove who was responsible.

Gives each agent an on-chain identity and records its actions as receipts anyone can re-verify.

Solidity / Monad / Agent Identity

Markdown docs ship with SEO problems nobody checks.

Combines rule-based checks with AI semantic analysis to audit Markdown for SEO.

SEO / Python

Two people want a deal but not the haggling.

Two agents negotiate on behalf of their humans, streamed live so both sides can watch the reasoning.

Multi-Agent / Streaming

MCP sessions are hard to debug after the fact.

Records every JSON-RPC message between client and server, then replays the session in a visual debugger.

MCP / JSON-RPC

Generated SQL looks right until it runs.

Writes dbt models from a real DataHub catalog, then proves the SQL is correct before you see it.

dbt / DataHub / Agents

Agent fleets overwrite each other's memory.

Serializable, auditable shared memory for agent fleets on CockroachDB.

CockroachDB / AWS / Agents

New AI benchmarks appear faster than anyone can track.

Collects and scores AI benchmarks every day, automatically.

Automation / Python