OpenAI Hackathon Finalist
| 86k+ | 1,000+ | Finalist |
|---|---|---|
| production queries studied | users reached at Wonka | OpenAI Hackathon |
| LLM routing research | shipped AI products | built under pressure |
Customer-facing AI products and integrations used by 1,000+ people.
I work across product, engineering, and deployment: building integrations, agent infrastructure, and features shaped directly by customer workflows.
TypeScript · Python · MCP · LLM infrastructure
A self-hosted voice-agent appliance for outbound sales teams.
Sales reps manage fleets of AI callers that qualify leads, update the CRM, and improve from each campaign. Every operator owns their SIP trunk, models, data, and costs instead of depending on a shared calling platform.
Bun · React · MCP · Pipecat · LiveKit SIP · Postgres
Multi-objective routing for production LLM systems.
Routes each query according to performance, cost, latency, and environmental impact instead of relying on a single default model.
Developed from my ENSAM engineering thesis and validated against 86k+ production queries.
Python · FastAPI · LLM evaluation · multi-objective optimization
OpenAI Hackathon finalist — safe phone actions for software agents.
A local MCP lets Codex and other compatible agents turn a task into a real, consented phone call. Classification, preview, confirmation, strict allowlists, and structured results are part of the action contract.
Python · MCP · Twilio · Deepgram Voice Agent
| Project | What it explores |
|---|---|
| Mirror | Local-first macOS workflow assistant that recognizes repeated work and prepares reviewable automation drafts without collecting screenshots, keystrokes, or page content. |
Boring · Chiant |
Computer-vision parking assistant combining control-vehicle detection, geofencing, and programmatic parking payment. |
| MCP integrations | Connectors for Bowimi, Odoo, and other business software so agents can safely act inside operational workflows. |
| Gauntlet | Parallel AI-agent harness for evidence-backed QA, code review, security audits, and implementation loops. |
| Forge | Local-first Life OS for morning execution, habits, training, focus, and personal progression. |
| Career OS | Native Mac workspace for application tracking, CV tailoring, interview preparation, and a live copilot. |
See the full portfolio at iamgabriel.xyz.
- LLM routing and evaluation;
- agent infrastructure that remains observable in production;
- integrations grounded in real operational workflows;
- working directly with users to turn vague needs into deployed systems.
Growing toward Forward Deployed Engineering: deep enough in the stack to build the system, close enough to users to know which system should be built.