// PROJECTS
Four projects, end to end.
Each project ships through the full pipeline: training data, model, orchestration, evaluation, and deployment. Click any card for the architecture and the numbers.
Selected projects
iRacing AI Race Strategist
ShippedEval score 29.7 → 78.5 · hallucinations 48/55 → 0/55 · 800–1100 ms end-to-end
Fine-tuned Llama 3.1 8B that calls race strategy in real time — and the eval harness that proved it works.
- Llama 3.1 8B
- QLoRA
- transformers / peft
- LM Studio
- GGUF q4_k_m
- Piper TTS
testpilot — Self-Repairing QA Agent
Shipped113/113 tests green · verified demo: UI drift auto-repaired, real regression refused
Turns plain-English QA stories into Playwright tests that diagnose their own failures — and refuse to weaken themselves when the product is what broke.
- TypeScript
- Playwright
- Vitest
- React + Vite
- SSE
- MCP (GitHub · Jira)
AWS Bedrock Agentic Shopping Assistant
LiveLive at shop.higuera.io · full AWS infra (Lambda, DynamoDB, API Gateway, S3, IAM) in Terraform
Autonomous shopping agent on AWS Bedrock — natural-language search, tool-calling, fully Terraform-provisioned.
- AWS Bedrock Agents
- Lambda
- DynamoDB
- API Gateway
- S3
- IAM
Titanic Historical RAG
Live3,400+ pages · 167 witnesses across both inquiries · Hit@5 70% · p50 search 276 ms
A RAG search engine over both 1912 Titanic inquiries that surfaces contradictions between witnesses instead of hiding them — including the same witness telling two different stories.
- OpenAI text-embedding-3-large
- Pinecone
- Claude Haiku 4.5
- FastAPI
- DynamoDB
- pymupdf