Rio de Janeiro, Brazil 🇧🇷 · Remote Worldwide 🌐
Building resilient backend microservices, high-throughput APIs, and production Generative AI systems.
12+ Years in IT • 6+ Years with Python • 3+ Years Production GenAI / Agents
FastAPI • PostgreSQL • pgvector • Docker • GCP • RAG • MCP • LangGraph
Backend: Python (3.10–3.13) · FastAPI · Pydantic v2 · SQLAlchemy · Elysia / Bun · REST APIs · Asyncio
AI & Agents: OpenAI API · Anthropic Claude · Tool Calling · MCP · LangGraph · Hybrid RAG · Local LLMs
Databases: PostgreSQL 16 · pgvector · Redis · SQLite · Supabase · Schema Modeling & Migrations
Infra & Ops: Docker · Linux (Debian/Ubuntu) · GCP (GCE, Cloud Run, Secret Manager, IAP) · Traefik v3 · CI/CD
Quality: Pytest · Ruff · Mypy · Clean Architecture · Observability & Structured Logging
- Architecture: Hybrid search microservice combining PostgreSQL + pgvector (HNSW) with Full-Text Search via Reciprocal Rank Fusion (RRF).
- Impact: Cut query retrieval latency from minutes to <200ms across millions of state records with 4.8x recall improvement.
- Stack:
Python 3.12·FastAPI·PostgreSQL·pgvector·Docker·Redis
- Architecture: Automated PDF ingestion and data extraction pipeline using LLM Tool Calling and Pydantic v2 validation with self-healing correction loops.
- Impact: Achieved 99.2% schema validation accuracy, reducing manual document data entry time by 85%.
- Stack:
Python·FastAPI·Pydantic v2·Claude/OpenAI APIs·Pytest
- Architecture: Dual-engine architecture (Bun perimeter ingestion + FastAPI reasoning core) connected to messaging protocols (Evolution API for WhatsApp).
- Impact: Processes thousands of real-time webhooks with sub-1.2s agentic responses, handling automated qualification, dynamic pricing, and CRM sync with zero message loss.
- Stack:
Bun·FastAPI·PostgreSQL·Redis·Docker·Evolution API
- Architecture: Specialized vertical RAG engines with multi-stage retrieval, cross-encoder re-ranking, and deterministic legal/STEM citation guardrails.
- Impact: Reduced hallucination rates to <1.5% on rigorous multi-step legal reasoning benchmarks.
- Stack:
Python·RAG·LangGraph·pgvector·Hugging Face Datasets
┌──────────────────────┐ ┌──────────────────────┐
│ External Clients │ │ Webhooks / Gateway │
│ (React/Next.js/Apps) │ │ (WhatsApp/Messaging) │
└──────────┬───────────┘ └──────────┬───────────┘
│ │
▼ ▼
┌────────────────────────────────────────────────────────┐
│ Traefik v3 Cloud Edge (TLS/SSL) │
└──────────────────────────┬─────────────────────────────┘
│
┌───────────────┴───────────────┐
▼ ▼
┌─────────────────────┐ ┌─────────────────────┐
│ FastAPI Services │ │ Bun/Elysia Gateway │
│ (Core Domain Logic)│ │ (High-Throughput In)│
└──────────┬──────────┘ └──────────┬──────────┘
│ │
├───────────────┬───────────────┘
▼ ▼
┌────────────────────┐ ┌─────────────────────────────────┐
│ PostgreSQL 16 │ │ Autonomous Agent Engine │
│ (pgvector / Rel.) │ │ (Tool Calling, MCP, Hybrid RAG) │
└────────────────────┘ └────────────────┬────────────────┘
│
▼
┌─────────────────────────────────┐
│ LLMs: OpenAI / Claude / Local │
└─────────────────────────────────┘
- Email: moises.costa12345@gmail.com
- LinkedIn: linkedin.com/in/moises-costa-rj
- Time Zone: Rio de Janeiro, Brazil (UTC-3 / EDT-1) — Smooth business hours overlap with US (EST/PST) and European (CET) teams.