Data Scientist & AI Engineer · Madrid, Spain
I design, build and operate production AI systems end to end — architecture, models, infrastructure, and the on-call that comes with it. Nearly four years shipping ML, NLP and Generative AI for clients in energy, insurance, SaaS, legal tech and medical imaging.
Everything below runs on infrastructure I host and maintain myself, which is the honest test of whether the engineering holds up.
🌐 manapple.dev · en español | 📄 CV | 💼 Talks, mentoring & consulting | ✉️ christopher.manzano@manapple.dev
Reinforcement Learning in Portfolio Management — Bachelor Thesis, Universidad Carlos III de Madrid, 2023
A Deep Reinforcement Learning agent that manages a portfolio end to end — deciding not just how to rebalance, but whether to rebalance at all. Soft Actor-Critic (SACv2) over a frozen Temporal Convolutional Network encoder, benchmarked against equal-weight, minimum-volatility and maximum-Sharpe portfolios on S&P 500 data split across 1986–2023.
Read the thesis · Source · Overview and case studies
| Project | What it does |
|---|---|
| warren | LangGraph multi-agent investment analyst. Runs quantitative, fundamental and macro analysis, then uses LLMs to synthesise a buy/sell thesis with entry, take-profit and stop-loss — but only when a credible thesis exists. |
| gollum | Auction tracker that ranks listings on quality-for-price rather than lowest price, with self-hosted email alerts and on-demand LLM vision on the listing photos. |
| manapple.dev | The site itself: Astro, bilingual, self-hosted behind my own Cloudflare tunnel and Coolify stack. |
Write-ups for each are at manapple.dev/writing.
I take on a limited amount of outside work in Spanish or English:
- Talks & workshops — applied AI built from systems actually running in production, for teams, universities and events.
- Mentoring — one-to-one for data scientists and AI engineers making the jump from analysis to engineering.
- Freelance consulting — architecture, LLM and agentic pipelines, private LLM deployment, and rescuing proofs of concept that only work on a laptop.
Python PyTorch LangGraph LangChain Hugging Face FastAPI scikit-learn
AWS Databricks Snowflake BigQuery MLflow Docker PostgreSQL R
- reinforcement_learning_in_finance — the thesis codebase: DRL trading agents, data pipeline and unified backtesting.
- mlflow_packaging_samples — worked examples of the different ways to package an ML model with MLflow.
- latent_dirichlet_allocation-case-study — applied LDA topic modelling, with the knitted report published.
- death_penalty_nlp — Naive Bayes stance detection on a hand-labelled Twitter corpus.
- face-recognizer_with_fda_pca_knn — PCA + Fisher discriminant + k-NN face recognition, built from the statistics up.
- born2beroot — the 42 Madrid system-administration write-up, still the most-starred thing I've published.