Been into tech since I was 11. Started by rooting Android phones and compiling custom kernels just because I needed to know how things worked under the hood. Never really stopped.

These days I build production LLM systems and agent architectures. Multi-agent orchestration, RAG pipelines, AI systems and MCP servers.

First to know about any interesting hardware or framework release, always have been.

Currently exploring: Advanced AI agent patterns, TurboQuant, Rust for systems work, and whatever drops next.

Outside Engineering:

Barça since childhood. Watched the trebles, cheered at the 6-1 comeback, had to somehow process Messi's departure. Still watching every game, still believing in the rebuild.

Mercedes in F1. Stayed through the porpoising era, stayed through the W14 being completely undriveable. Kimi for the win. We move.

Python LangGraph FastAPI Azure OpenAI

Multi-agent system that generates, tests, and refactors code autonomously. Four specialized agents orchestrated via LangGraph using ReAct and Plan-and-Execute patterns with tool calling, memory, and reflection. Real-time streaming via WebSockets with quality scoring and regression gates.

Python LangChain Chroma Pinecone DuckDB

AI assistant that combines document search with SQL analytics — hybrid retrieval using RAG for unstructured data (Chroma) and natural language to SQL for structured data (DuckDB). GPT-4 query classification, Cohere reranking, and fallback mechanisms when retrieval confidence is low.

Next.js TypeScript Python MCP Gemini

Resume management tool I built because I got tired of managing 10 different resume versions manually. Git-based version control, DOCX parsing via JSON-RPC bridge, automated PDF generation with Puppeteer, and MCP integration for AI-assisted editing.

Python PyTorch Graph Neural Networks Drug Discovery

Two hypergraph neural network approaches for molecular toxicity prediction. K-clique model uses dense subgraphs with heterogeneous message passing, functional group model uses SMARTS pattern matching. Achieved 0.894 ROC AUC on Tox21 androgen receptor dataset. Includes complete research paper.

C TCP/IP Sockets Systems Programming

Four-node distributed file system in pure C with custom TCP protocol. Each server manages independent namespace with upload, download, delete, and tar operations. Client routes requests to appropriate nodes. Built from scratch without frameworks to understand low-level networking.