Rui S.

I'm a senior AI engineer with a decade of experience across applied AI, software engineering, and tech consulting. I work with private equity firms, venture capital funds, family offices, and asset managers — both on the AI systems that automate their internal operations and as a forward-deployed engineer inside their portfolio companies. I spent four years inside a €600M B2B software fund, where I was the lead AI engineer responsible for shipping production AI systems that automated core fund workflows. Before that, I led digital transformation projects for large commercial banks at Deloitte Digital and KPMG. 𝐖𝐇𝐀𝐓 𝐈 𝐁𝐔𝐈𝐋𝐃 𝐈𝐧𝐭𝐞𝐫𝐧𝐚𝐥 𝐀𝐈 𝐟𝐨𝐫 𝐟𝐮𝐧𝐝𝐬 𝐚𝐧𝐝 𝐚𝐬𝐬𝐞𝐭 𝐦𝐚𝐧𝐚𝐠𝐞𝐫𝐬. Unified deal intelligence over fragmented sources (decks, websites, CRM, email, transcripts), automated deal sourcing with thesis-aligned filtering, CRM action layers via MCP, and standardized investment memos generated from the corpus. These are the systems I developed and ran for 4 years inside an active investment team. 𝐎𝐩𝐞𝐫𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐀𝐈 𝐟𝐨𝐫 𝐩𝐨𝐫𝐭𝐟𝐨𝐥𝐢𝐨 𝐜𝐨𝐦𝐩𝐚𝐧𝐢𝐞𝐬. The bottlenecks I worked on inside funds show up almost identically inside portfolio companies — most often in logistics, manufacturing, procurement, and adjacent industries. The operational drag is the same: document-heavy processes where structured outputs turn unstructured text into useful data, and fragmented internal knowledge that needs RAG and function calling to be usable from inside chat or ERP-style interfaces. I've delivered this work for regulated clients in financial services, so role-based access control, PII redaction, audit trails, and tenant isolation are baked into how I build. 𝐒𝐄𝐑𝐕𝐈𝐂𝐄𝐒 I deliver AI projects end-to-end, without hand-offs between strategy and implementation. • 𝐒𝐭𝐫𝐚𝐭𝐞𝐠𝐲. Identify the AI opportunities that actually move a business metric. The job is to find the 20% that drives 80% of the outcomes. • 𝐈𝐦𝐩𝐥𝐞𝐦𝐞𝐧𝐭𝐚𝐭𝐢𝐨𝐧. Architect and ship the system. Production-grade, instrumented, evaluated, and integrated with existing infrastructure. Open-source-first to avoid vendor lock-in. Deployable on-prem or in the cloud depending on security and infrastructure constraints. • 𝐄𝐧𝐚𝐛𝐥𝐞𝐦𝐞𝐧𝐭. Transfer capability to the in-house team so they can operate, extend, and own what we built. 𝐌𝐘 𝐋𝐄𝐍𝐒 𝐎𝐍 𝐀𝐈 • 𝐏𝐫𝐨𝐜𝐞𝐬𝐬 𝐨𝐯𝐞𝐫 𝐭𝐨𝐨𝐥𝐬. AI engineering is still 80% systems engineering. The model is one of many moving pieces and rarely the right one to change first. • 𝐏𝐫𝐨𝐦𝐩𝐭 𝐚𝐧𝐝 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞 𝐛𝐞𝐟𝐨𝐫𝐞 𝐟𝐢𝐧𝐞-𝐭𝐮𝐧𝐢𝐧𝐠. Most teams reach for fine-tuning too early. The leverage almost always sits earlier. • 𝐀𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐮𝐫𝐞 𝐨𝐯𝐞𝐫 𝐜𝐨𝐦𝐩𝐨𝐧𝐞𝐧𝐭𝐬. When the system is failing, the fix is usually in the architecture, not in the components. Better pipeline architecture beats swapping models or databases. • 𝐄𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧-𝐜𝐞𝐧𝐭𝐫𝐢𝐜 𝐝𝐞𝐯𝐞𝐥𝐨𝐩𝐦𝐞𝐧𝐭. Evals are how you turn a non-deterministic system into a reliable one. Without them, every prompt change is a coin flip and every regression goes undetected. • 𝐒𝐤𝐞𝐩𝐭𝐢𝐜𝐚𝐥 𝐨𝐟 𝐟𝐮𝐥𝐥𝐲 𝐚𝐠𝐞𝐧𝐭𝐢𝐜 𝐬𝐲𝐬𝐭𝐞𝐦𝐬. A deterministic pipeline with LLM calls at specific steps beats a fully agentic system for most production problems. Map the process before adopting agentic flows. 𝐓𝐄𝐂𝐇 𝐒𝐓𝐀𝐂𝐊 • 𝐋𝐋𝐌 𝐨𝐫𝐜𝐡𝐞𝐬𝐭𝐫𝐚𝐭𝐢𝐨𝐧: LangChain, LangGraph, CrewAI, Pydantic AI, LangFuse, LiteLLM, OpenAI and Anthropic SDKs. • 𝐒𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞𝐝 𝐨𝐮𝐭𝐩𝐮𝐭𝐬 𝐚𝐧𝐝 𝐯𝐚𝐥𝐢𝐝𝐚𝐭𝐢𝐨𝐧: Pydantic schemas as the contract between every LLM call and the rest of the system. • 𝐑𝐞𝐭𝐫𝐢𝐞𝐯𝐚𝐥 𝐚𝐧𝐝 𝐯𝐞𝐜𝐭𝐨𝐫 𝐬𝐭𝐨𝐫𝐞𝐬: PostgreSQL with pgvector by default. Qdrant, Weaviate, Pinecone when the engagement warrants a dedicated store. • 𝐃𝐚𝐭𝐚𝐛𝐚𝐬𝐞: PostgreSQL, MySQL, Supabase. Alembic for migrations, SQLAlchemy when an ORM is needed. • 𝐀𝐏𝐈 𝐚𝐧𝐝 𝐚𝐬𝐲𝐧𝐜: FastAPI, Flask, Celery. • 𝐃𝐒 / 𝐌𝐋 / 𝐃𝐋: Pandas, NumPy, scikit-learn, XGBoost, TensorFlow, MLflow, Matplotlib. • 𝐎𝐩𝐬 𝐚𝐧𝐝 𝐜𝐥𝐨𝐮𝐝: Git, GitHub, Docker, uv for Python packaging, Streamlit for internal UIs. GCP, AWS, Azure, DigitalOcean, Heroku — I follow client preference. • 𝐈𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐢𝐨𝐧𝐬: Slack, WhatsApp, Telegram, Facebook Messenger, Instagram for conversational front-ends. MCP for tool-use and CRM action layers.