Work

Telos Internship2026 to now AI candidate sourcing in daily production at a 140-person IT staffing firm, serving about 80 open requisitions a month. Ranks candidates on placement history rather than keyword overlap; I own the retrieval and ranking end to end. FastAPI · Postgres · Pinecone · React · JobDiva · PDL Airlock Shipped2026 · hackathon Agents will get fooled. Getting fooled shouldn’t move money. Zero unauthorized payments across 39 injection attacks. A finance team of AI agents reads invoices from email, disputes overcharges with other companies’ agents, and verifies bank changes against the vendor’s phone on file. FastAPI · Claude · Primitive · Inkbox · Stripe Connect · MCP What agents don’t test Research2026 · solo Opus 5.5 solves 4 of 9 runs on the third version of this RL coding task. Every failing run fuzzed against a reference implementation, reported a match, and missed exactly the rules its fuzzer never generated; three pilots in a real codebase came back clean and surfaced a bug now fixed upstream. Python · Docker · Claude Code · 33 graded runs · 20+ grader bugs logged · pelican#3621 vaudit Research2026 · solo A correct emulator scored below a blank screen. Three of five published grading strategies rank a blank screen above a correct implementation. Five CHIP-8 ROMs, each isolating one platform quirk and pre-registered before its bytes existed; a follow-up falsified 4 of its own 14 predictions. Python 3.13 · NumPy · pytest · CI on every push · 6 upstream bugs + 1 PR filed Goodhart Research2026 · hackathon Hardens RL verifiers against reward hacking: a red-team swarm finds answers that pass the grader but fail an independent oracle, then a green team patches the grader. Reports a before/after robustness number against held-out ground truth. Finalist, HUD × Y Combinator. Python 3.12 · Anthropic SDK · FastAPI + WebSocket · pytest Envoy Shipped2026 · hackathon A voice agent you hand a real errand to: it calls the venue, negotiates with a live host, and books the table. After one breach it writes itself a guardrail that took credential-leak refusal from 0/5 to 5/5 under an independent judge. Pipecat · NVIDIA Nemotron · Gradium TTS · Twilio · Cekura Hindsight Shipped2026 · hackathon Grades your past predictions against what actually happened and corrects future advice for your blind spots. The concept shipped into GBrain two days later as the “Hindsight calibration wave”; 4th place, GStack × GBrain. TypeScript · Express · Chrome extension · GBrain skills Yaad Shipped2026 · hackathon A Hindi/English voice companion for people with early dementia. Answers come from a retrieved memory graph or not at all; below a confidence threshold it says it doesn’t know. LiveKit · Pipecat · MiniMax TTS · Moss · FastAPI · Supabase Vibecheck Shipped2026 Paste a GitHub repo, get a plain-English security report, then one click opens a PR fixing the top issues. Semgrep, Gitleaks and OSV run inside a sandboxed engine that never executes the target code; the LLM only translates findings and writes patches. Next.js · FastAPI · Semgrep · Gitleaks · OSV · Claude · Upstash Strata Shipped2026 · v1.0 Field guidance for plumbing contractors, deployed with 3 pilot contractors and cutting per-visit documentation time by 20%. Camera diagnosis and voice input, with full offline operation from a local FTS5 cache. React Native (Expo) · FastAPI · GPT-4o Vision · Pinecone · SQLite FTS5 pipecat-firewall Shipped · PyPI2026 A drop-in guardrail for Pipecat voice agents, published on PyPI. Blocks or redacts prompt injection and PII extraction before the model runs, with deterministic regex detectors, no API key and no added latency. Python · published on PyPI · MIT · CI Sentry Shipped2026 · hackathon A voice agent that leaks a card number under social engineering, then rewrites its guardrails mid-call and refuses the same attack. The README states which parts are live and which are replayed. NVIDIA Nemotron · Pipecat · Daily WebRTC · Deepgram · ElevenLabs RepairSync ↗ Shipped2026 · hackathon A neutral AI referee for auto-repair disputes. Audits collision estimates against live parts pricing, flags overcharges and skipped safety steps, and issues a report all sides trust. TypeScript · Convex · web dashboard Signal ↗ Shipped2026 An AI-curated daily newsletter. Fetches RSS, uses Claude to select and summarize, sends via Resend, archives to Supabase. The pipeline runs on a GitHub Actions cron; the Next.js dashboard only manages it. Node · Claude · Resend · Supabase · GitHub Actions · Next.js InternTracker ↗ Shipped2026 Internship application tracker with AI company research and email parsing, paste any recruiting email and it extracts company, role, status, and deadline, then updates the board. Next.js · Supabase · Claude Meeting Prep ↗ Shipped2026 Generates investor briefing documents ahead of meetings by pulling live web search, CRM notes, and fund thesis context into one structured brief. Built against M13's workflow. FastAPI · Tavily · OpenAI · FAISS · Airtable · React Kalshi vs. options IV Research2026 Kalshi unemployment markets Granger-cause VIX at a two-day lead (p = 0.024, 222 days). CPI markets through the same pipeline show nothing, which is the null control. Python · Kalshi API · yfinance · statsmodels Transaction Pattern Analyzer ↗ Research2026 Anomaly detection on business transaction data, 10 engineered contextual features, three detection methods compared, Isolation Forest against rule-based scoring. Deliberately a prototype: the notebook is honest that interpretability mattered more than model complexity. Jupyter · scikit-learn · pandas Trialsignal ↗ Dead2026 Alternative-data platform to forecast consumer class actions before they file. The thesis was good and the data acquisition was the whole problem. I built the front end before I could prove I could get the signal. Wrong order. Next.js · shadcn/ui · TypeScript Tab Memory Agent Dead2026 A Chrome extension that gave your browser persistent memory of intent, not just URLs. Passive capture of every page you spent 15+ seconds on. Killed it: the privacy cost of always-on capture was not worth the recall benefit, and I did not want to run that on myself. Chrome extension · mem0 · Cartesia · Python SpendLens ↗ Dead2026 Upload a bank CSV, ask questions in plain English, get charts. Worked fine. Died because it is a feature, not a product, every bank and budgeting app ships this now. Python · Streamlit · GPT-4o-mini Referral Hunter Dead2026 · private Automated finding warm referral paths into companies. Shelved, the interesting half of the problem turned into Telos, which is the same matching question pointed at a market that pays for it. Python P-oker ↗ Dead2026 Agentic poker AI, simplified environment, self-play agents, opponent modeling, an EV and win-rate dashboard. A good excuse to learn self-play; never got past the toy environment. Python · Supabase DoubleJack ↗ Dead2026 A class project, finished and left alone. , Pantry Tracker & Flashcard SaaS ↗ Dead2024 The two Next.js tutorial projects everybody builds. Including them because this is where I started, and pretending otherwise would be silly. Next.js · Firebase NeetCode submissions ↗ Ongoing2026 Not a project. Just the receipts. Python