Finished game week at @GauntletAI and shipped a full kart racer in the browser...
Rollback netcode, 0ms local input delay
Drift boosts, shells, oil slicks
Grand Prix cups + battle mode
Ghost time trials, un-forgeable leaderboards
mkultra.jagger.lol/
Wrapping up this week's @gauntletai project: took an open-source issue tracker and turned it into a developer platform. OAuth 2.0 with PKCE + device flow, webhooks with retries + replay, and a TypeScript SDK + CLI. Then rewrote last week's AI agent to authenticate thru the SDK.
Halfway through @gauntletai and wrapping up this week’s project: an AI agent for an open source issue tracking tool.
Fleet is an autonomous LangGraph agent that detects sprint drift, duplicate tickets, and sprint plans without clear goals. Under the hood: scoped context fetches, structured LLM outputs, and HITL writes.
Week 4 @gauntletai wrapped up: spent the week hardening Ship, a real-time collaborative workspace.
Ran it through security, performance, and accessibility audits, fixed what surfaced, and deployed, along side Probe, a security scanner that hammers the deployment across auth, websockets, inputs, dependencies, rate limits, and ranks every finding.
Finishing week 3 @gauntletai with a fun project: built an adversarial security platform for attacking and hardening AI agents.
Used LangGraph to orchestrate model-assisted agents that choose threat surfaces, generate attacks, run them against a live target, and judge the outcomes.
Found exploits become regression tests to catch the bug if it ever comes back.
Wrapping up Week 2 at @gauntletai shipping some new features into a legacy PHP EMR
📄 Document ingestion with bbox citations. Drop a lab PDF or intake form into a patient's chart, ask the agent about it, and every cited fact deep-links back to the exact pixel region on the source page
🖥️ Brand-new Next.js frontend. Spun up a modern dashboard and bridged it into the legacy PHP UI
📚 Hybrid clinical-guideline RAG. BM25 + Cohere embed + rerank over USPSTF/CDC guidelines, so the agent can cite real recommendations
I’m wrapping up Week 1 at @gauntletai!
My first challenge: take a legacy PHP electronic medical records system and build an AI chat agent that can interact with patient data quickly.
The goal: let physicians talk with patient data instead of fumbling through an old UI.
Built with a multi-turn LangGraph agent + Langfuse E2E tracing/observability.




