Full Stack Developer who ships AI into production, not into demos. Four of the projects below run every day at real businesses — an autonomous WhatsApp agent that serves live customers, an analytics app, a mobile POS, and a humanitarian real-time feed.

I measure what I ship. The retrieval pipeline behind my WhatsApp agent is benchmarked against 320 real customer queries (76.9% exact recall, 6.3% failure rate), and I redesigned the embedding input to lift the vector layer's signal-to-noise 10x.

Coming from an Operations Management background (where I led a team of 7 and cut inventory discrepancies by ~30%), I approach engineering problem-first: I've seen the bottlenecks from the inside and built the software to fix them.

• 🌍 Based in: Venezuela · Available for 100% Remote Roles
• 🗣️ Languages: Spanish (Native) · English (B2 – Upper Intermediate)
• 🧠 Focus: RAG & Vector Search · AI Agents & Tool Calling · LLM Evaluation · Full-Stack SaaS · React Native
• 📫 Reach me at: gusreyes_dev@outlook.com · LinkedIn

AI & Retrieval

Frontend & Mobile

Backend & Data

Engineering Practice

Software in real production use, with a focus on AI integration, secure data access, and measured results.

An event-driven WhatsApp agent serving real customers daily at Ferretería El Serrucho: answers inventory questions, generates exact quotes from live POS pricing, transcribes voice notes, and escalates to staff when a human is needed.

• Tech: n8n · WAHA · OpenAI (via OpenRouter) · text-embedding-3-small + pgvector · Groq Whisper · Supabase · PostgreSQL · Docker.
• Highlights: 5-layer retrieval pipeline over a 7,000+ SKU catalog (lexical & measurement parsing → 3.5k-term catalog dictionary → pg_trgm fuzzy matching → vector search → LLM semantic rescue), resolving the common query in ~900 ms at zero marginal cost.
• Measured: benchmarked on 320 real customer queries — 76.9% exact recall, 6.3% failure rate. Redesigning the embedding input (description + category + colloquialisms) lifted the vector layer's signal-to-noise 10x, adding +7 correct matches over lexical-only retrieval.
• Production guards: read-only DB connection for safe pricing · LLM output sanitization · rate limiting · CI sync tests.

A mobile + PWA analytics dashboard in daily use at the store, shipped as an Android app (EAS) and an installable PWA from a single codebase via react-native-web.

• Tech: Expo · React Native · TypeScript · Supabase · PostgreSQL · Python · Google Gemini.
• Highlights: closed-loop writeback engine bridging an on-premise POS to the cloud through a Python watchdog · Gemini anomaly detection in Supabase Edge Functions · RLS role-approval gates · the staff queue that resolves the AI agent's escalations · virtualized inventory list.

A tablet-first cross-platform POS actively used by a paying café client, built for small businesses with unstable internet.

• Tech: React Native · Expo (SDK 54) · TypeScript · Supabase · PostgreSQL.
• Highlights: offline-first inventory · real-time sync · multi-tenant access via Row Level Security · barcode scanning · branded PDF invoices · date-range financial dashboards with transaction cancellation and automatic stock restoration.

A verified real-time information feed built with the Techeasy LATAM community during the Venezuela earthquake, to counter misinformation when it mattered most. 73 commits contributed.

• Tech: Next.js · TypeScript · Supabase · Claude API · PWA.
• Highlights: AI-assisted verification of incoming reports · installable PWA for low-connectivity conditions · built and shipped under real time pressure with a distributed team.

A full-stack FinTech app that removes the friction of group expenses through AI data extraction and optimized settlements.

• Tech: React 18 · Flask · PostgreSQL · Supabase · Google Gemini · PayPal · Vercel · Render.
• Highlights: heap-based settlement algorithm that cuts inter-party transactions by up to 50% · Gemini-powered receipt parsing · 30+ RESTful endpoints with JWT lifecycle, bcrypt and rate limiting · dual-cloud deploy via GitHub CI/CD.