DeepCode combines software architecture, industrial data engineering and generative AI for healthcare and pharmaceutical manufacturing. Work directly with Rodrigo Manso, a software professional since 2001 with international consulting experience.
$ deepcode deploy --target production ✓
Four practice areas, one standard: solid architecture first, real AI second. Nothing lands in your system without a clear path to production.
LLMs, RAG and agents integrated with your systems and your data — with evaluation, guardrails and cost under control. From pilot to product.
Design and evolution of microservices architectures: messaging, events, APIs and cloud. Systems that scale without becoming hostages of their own code.
Pipelines, modeling and AI-ready data delivery. Because no model is better than the data that feeds it.
Senior engineering to build what the market doesn't sell off the shelf — .NET/C#, Python and Node.js, across healthcare, government and international projects.
Long-term engagements, not drop-and-run projects. The cycle is always the same:
We map architecture, data and the real business problem — before talking about models. Sometimes the right answer isn't AI at all.
We define the architecture, contracts, evaluation and costs of what will be built. The pilot is born with its path to production already drawn.
We build, ship to production and stay: monitoring, continuous evaluation and evolution alongside your team.
DeepCode was founded by Rodrigo Manso, a solutions architect with 25 years of mission-critical systems — from public administration platforms used by major Brazilian cities (Curitiba, São Paulo, Fortaleza) to healthcare systems and international AI projects.
From machine learning research and industrial data engineering to retrieval pipelines, agents and AI-assisted software development.
Research, production delivery and independent experiments build on each other. Several strands of this journey continue in parallel.
At ICI, I led research and prototyping across machine learning, computer vision, IoT, chatbots and software agents for smart-city and public-administration use cases. My learning covered algorithms, data preparation, model training, tools and pipelines.
The innovation team I led won first place at the Sebrae Analytics Hackathon.
As a software architect, I helped transform a monolithic industrial data integration platform into decoupled readers, formatters, writers and schedulers. Messaging, state machines, chunking and bounded parallelism enabled terabyte-scale extractions in production on servers with 16 GB RAM.
In January 2025, I began my first AI-generated application implementation: a client telemedicine platform. I direct the architecture, review and validation, covering WebRTC calls, WebSocket signaling, Coturn STUN/TURN, participant roles, file sharing and administrative controls. Deployment and restrictive-network validation are in progress.
Built a client-specific RAG pipeline across six modules of a long-established software platform. Database artifacts, code analysis, Moodle content and business metadata provide context about functionality and relationships. Chroma stores the retrieval index; Qwen 2.5 Coder runs through Ollama. The knowledge base is refreshed weekly, without model fine-tuning.
Architect and develop an AI-assisted code-generation product using Claude Code in headless mode, GitHub/Azure DevOps integration and human-reviewed pull requests. An MCP server exposes project knowledge and work-item tools. In a client pilot, four AI-assisted changes were validated and deployed to production.
My independent projects extend this work into reproducible data and AI experiments. Data Intelligence includes logistic-regression training with chronological splits. Motorsport Engineering Lab combines a Python historical-data pipeline and a Rust telemetry historian; simulator integration and AI-assisted lap experiments remain future milestones.
Meet ERPrint, our modular platform for print businesses, and Javisei, our customer-service platform. Two products, built around everyday operations.
An eight-component engineering lab connecting public historical data and game telemetry. ETL implementation, UDP header parsing, edge filtering/buffering and distance-indexed lap-script prototypes are in progress. The Rust historian already has a local v1; the roadmap adds durable replay, comparative dashboards and controlled AI experiments. Integrated live operation and benchmarks remain pending.
A product concept in the requirements phase, proposing a social jukebox with videoke: AI vocal separation (source separation), word-level synced lyrics via forced alignment, real-time singing score (pitch + timing), collaborative QR-code queue and B2C→B2B monetization with licensing requirements still to be validated.
Tell us your context in two lines. The first conversation is an honest diagnosis — including when the answer is "not yet".
contato@deepcode.com.br