Alvaro Serero

Pi2 industrial research project, ESILV, with asset manager Ginjer-AM

Technical pipeline lead on a causal ML research project with asset manager Ginjer-AM: reduced 288 features to the 10 validated causal drivers of Bitcoin returns, then built a monthly exposure strategy (Sharpe 1.03 vs 1.00 buy-and-hold, max drawdown -49% vs -83%).

Read the case studyShowcase siteTeam repo on GitHub

AI platform detecting full public trash bins · EFREI Mastercamp

Built an AI platform (team of 5) classifying photos of public trash bins as full or empty. Fine-tuned a MobileNetV2 CNN to 85% held-out accuracy (94% precision) and deployed it via ONNX at 3x smaller model size.

Read the case study

Web Datamining & Semantics, ESILV · a Wikipedia case study

Built an end-to-end knowledge-graph pipeline: crawler, transformer NER, RDF/OWL ontology aligned to Wikidata (65k triples), PyKEEN embeddings, and NL-to-SPARQL RAG with self-repair.

Read the case studySource on GitHub

LLM-driven Python code generator and sandboxed runner, in Rust

Built an LLM-driven Python code generator and sandboxed runner in Rust (8k lines, 117 tests, Docker sandbox, Axum/HTMX dashboard, from-scratch RAG). Led the 4-person course version (90% of the code).

Read the case studySource on GitHub

Full MERN web application · Software Engineering, ESILV

Built the core of a MERN personal-finance app (6 models, 31 JWT-protected endpoints, CSV/Excel import with MD5 deduplication, budget email alerts, 93 tests): 88% of a 9.5k-line codebase, deployed on Render.

Read the case studyTeam repo on GitHub

Grade-adjusted pacing for trail races · personal project

Upload a course GPX and get a per-kilometre pacing plan that plans steep climbs as power-hikes, self-calibrates from your own recorded runs, and gives an honest finish range instead of a single number. Built on the Minetti energy-cost model with a real-world descent correction. Runs entirely in the browser; your GPX never leaves your device.

Live appSource on GitHub

Information Retrieval & NLP, ESILV · team of 2

End-to-end NLP pipeline on 24k French insurance reviews: spell correction, LDA/LSA topic modelling, Word2Vec/GloVe embeddings, 9 rating and sentiment classifiers (TF-IDF+LR, CNN, BiLSTM, BERT, BART), SHAP, and a 6-page Streamlit app with RAG and QA.

Team repo on GitHub

Information Retrieval & NLP, ESILV · team of 2

Built a place recommender that uses review text only (153k reviews, 1,835 places). Tuned TF-IDF cut ranking error 40% vs BM25, and dense (MiniLM) and hybrid fusion retrieval were benchmarked in a 12-configuration ablation.

Team repo on GitHub

Containerised microservices app · Containerization Technologies, ESILV · team of 4

Co-built a Dockerised microservices app (Nginx, 2 replicated FastAPI services, isolated PostgreSQL databases, Redis Pub/Sub chat, 4 networks). Owned the Users API and the Trivy-gated Gitea CI/CD pipeline to Docker Hub.

Machine Learning, ESILV · team of 4, repo owner

Fraud detection on 285k Kaggle transactions (0.17% fraud): benchmarked Decision Tree, Bagging, Random Forest and XGBoost, raising fraud-class F1 from 0.67 to 0.84, and designed the cost-based decision-threshold analysis.

Source on GitHub

ESILV hackathon, 2 days · team of 5, repo owner

Built and deployed a 5-page Streamlit "Trusted AI" HR dashboard (SHAP explanations, GDPR anonymisation, 4/5-rule fairness audit, CodeCarbon frugality comparison) in a 2-day ESILV hackathon. Owned the repo, the docs and the live deployment.

Source on GitHub