AI Solution Architect & Full-Stack Engineer based in NYC.
I build AI-powered business systems end to end, from agent orchestration and backend integrations to the interfaces and workflows people actually use.
With a background as a founder and software engineer, I focus on solving real business problems through thoughtful architecture, practical AI, and complete software products.
🤖 Odoo AI Agent Production-Grade Voice-Enabled ERP Assistant → live demo
A voice-enabled AI agent that connects natural language to real business workflows in Odoo ERP.
Built as a multi-tenant SaaS application, combining a Next.js frontend with a FastAPI backend orchestrating a 21-node LangGraph agent.
- Agent architecture: Keyword-first routing, LLM fallback, query planning, validation and execution.
- Controlled actions: Entity resolution, clarification workflows and confirmation-gated CRUD.
- Production-oriented engineering: Multi-tenant authentication, role-based access control, SSE streaming, computed analytics, OCR, exports and Stripe billing.
- Voice interface: Speech-to-Text (Groq Whisper) + Text-to-Speech (Kokoro ONNX) with synchronized streaming
- Quality engineering: 2,000+ unit/characterization tests and 180+ documented end-to-end evaluation scenarios.
Stack: Next.js · React · FastAPI · LangGraph · PostgreSQL (Checkpoints) · Odoo XML-RPC
🎯 TalentGraph AI-Powered Voice Recruitment Platform → live demo
A full-stack recruitment platform built around a LangGraph agent, combining candidate screening, recruiter workflows, resume analysis, and conversational AI.
- Agent orchestration: 34 specialized nodes, conditional routing, and optimized fast paths.
- Multi-model architecture: OpenAI and Anthropic models for different tasks.
- Workflow automation: MCP integrations for Google Calendar and Gmail.
- Real-time experience: Streaming conversations, voice interaction, and PostgreSQL-backed agent state.
Stack: Angular 19 · FastAPI · LangGraph · PostgreSQL · MCP
📚 AI-Trainer-Teacher Intelligent RAG-powered Educational Platform → live demo
An AI-powered educational platform combining document-grounded assistance, adaptive assessments, and course management.
- Context-aware RAG: PDF retrieval with section-specific filtering and question reformulation.
- Adaptive assessments: Timed exams, automatic saving, and session recovery.
- Learning management: Administrative dashboards for courses, educational content, exams, and student progress.
Stack: Angular 19 · FastAPI · LangChain · Firebase · Vector Search
An exploration of bilingual Spanish/English speech recognition in noisy restaurant and drive-thru environments, using noise diagnostics, model adaptation, and evaluation pipelines.
🍔 Menu Detector — Deterministic Speech-to-Order Interactive Demo
A confidence-aware pipeline that transforms noisy speech-to-text output into structured restaurant orders validated against a known menu. It handles corrections, bilingual phrasing, and ambiguous items without silently guessing. The Menu Detector implementation is private.
Also on the shelf: LangGraph agent templates (6 production-ready agentic patterns) · OAuth 2.0 & JWT deep-dives (PKCE, Auth Code, real-world API integrations).
| Area | Technologies |
|---|---|
| Agentic AI | LangGraph, LangChain, MCP, tool orchestration, conditional routing |
| AI Applications | RAG, model routing, STT/TTS, document retrieval |
| Backend | Python, FastAPI, Node.js, REST APIs, SSE streaming |
| Frontend | Angular, React, Next.js, TypeScript |
| Data & Infrastructure | PostgreSQL, Redis, MongoDB, Firebase, Docker |
| Integrations | ERP systems, Stripe, Google Calendar, Gmail, OAuth 2.0 |
- 📍 Based in New York City.
- 🚀 Former founder with a product-oriented approach to engineering.
- 🌎 Fluent in English, French, and Spanish.
- 🇺🇸 Authorized to work in the U.S. without sponsorship.
- 🧠 Interested in agentic engineering, system design, and building software that solves real problems.
📧 info@mart-in.us · LinkedIn · Portfolio · Certifications
"The quick shall inherit the earth. I ship clean, performant code every week — measured, not assumed."