Jaime Berdejo Sánchez

Data & AI Engineer

Barcelona, Spain

+34 680 303 293 jaimeberdejo1902@gmail.com
Jaime Berdejo Sánchez

PROFILE

Computer Science graduate with a dual background in Business Administration and a Master's in AI Applied to Financial Markets, combining engineering depth with quantitative and business understanding. Hands-on experience building production ETL pipelines on AWS and developing LLM-based systems—from RAG and agentic applications to data-driven analytics platforms. I take projects from concept to production, turning complex problems into scalable, intelligent tools that deliver measurable impact.

EDUCATION

Master's Degree in Artificial Intelligence Applied to Financial Markets

University of Barcelona

Dual Bachelor's Degree in Computer Science & Business Administration

University of Granada

PROFESSIONAL EXPERIENCE

Data Engineer — Indra Sistemas
  • Data Warehouse Maintenance & Integrity: Managed the division's Data Warehouse, monitoring internal databases to detect and resolve inconsistencies, and maintaining documentation that kept critical data accessible for the analytics team.
  • ETL Pipeline Development: Built and optimized data pipelines in Java and Python on AWS Glue and Apache Airflow, automating data flows that improved processing efficiency and reduced manual intervention in the reporting cycle.
  • Cross-Functional Collaboration: Partnered with data analysts to translate business requirements into robust technical implementations aligned with the team's data infrastructure roadmap.
IT Intern — Grupo Bersan
  • Technical Support & Web Maintenance: Resolved technical tickets for web change requests and IT operations, keeping internal tools and web platforms fully operational.
  • Domain & Infrastructure Administration: Administered company domains and DNS configuration, ensuring proper record propagation and security protocols and contributing to 100% service uptime.

TECHNICAL SKILLS

Data Engineering & Cloud: AWS (Glue, S3, Lambda, EC2, EMR, IAM, SageMaker), Apache Airflow, dbt, PySpark, ETL/ELT automation, SQL (PostgreSQL, MySQL), NoSQL (MongoDB), Data Warehousing (Snowflake, Redshift, BigQuery), Schema design & optimization.

AI Engineering & GenAI: LangChain, LangGraph, Multi-agent system design, LLM integration (OpenAI, Claude, Gemini, Groq), RAG pipelines, Prompt engineering, Vector databases (pgvector).

Automation & Agentic Workflows: Python automation & scripting, n8n (low-code workflow automation), autonomous & multi-agent systems, AI agent orchestration, task scheduling (cron), webhooks & API integrations, end-to-end process automation.

Programming & Backend: Python (Advanced), Java, SQL, Bash, FastAPI, Django, REST API design, Git (GitHub/GitLab), Docker, CI/CD (GitHub Actions).

Data Science & Analytics: Pandas, NumPy, scikit-learn, XGBoost, PyTorch, Statistical modeling, Financial data analysis, Data visualization (Power BI, Tableau, Matplotlib).

LANGUAGES

Spanish: Native English: Fluent (C1 — Advanced)

SELECTED PROJECTS

FiscalPilot — Ongoing

Stack: Python, FastAPI, LangGraph, Claude API, Supabase (Postgres + pgvector), Next.js, PSD2 (GoCardless)

Agentic SaaS acting as a virtual CFO for Spanish freelancers: connects to bank accounts via PSD2, automatically reserves taxes (IVA/IRPF), classifies expenses, and proposes investment of idle cash—all under human-in-the-loop approval and full auditability (GDPR, EU AI Act, MiFID II compliant). Built on a deterministic tax engine with an agentic LangGraph layer above for reasoning.

RAG for European Financial Regulation — Master's Thesis (TFM) — Ongoing

Stack: Python, LangChain, vector databases, sentence-transformers, RAGAS, LLM APIs

Empirical study designing and evaluating Retrieval-Augmented Generation systems applied to the full hierarchy of European financial regulation (EU Regulations, RTS, ESMA Q&As, CNMV criteria). Building a modular RAG pipeline, releasing a public benchmark of validated reference Q&A pairs, and benchmarking chunking strategies, embedding models, and generator LLMs across standard retrieval metrics (Recall@k, MRR) and generation metrics (faithfulness, answer relevance) following the RAGAS framework and LegalBench-RAG methodology.

AutoVideo

Stack: Python, Typer, Claude API (Anthropic), ElevenLabs, Playwright, Jinja2, FFmpeg, WhisperX, Pydantic v2, Docker

Tool that fully automates the video-creation workflow: from a list of bullet points and a target duration, it produces a finished narrated slide video with zero manual editing. Orchestrates an end-to-end pipeline (LLM storyboard design → HTML/CSS slides rendered to PNG via Playwright → duration-calibrated voice-over script → ElevenLabs TTS or mic recording aligned with WhisperX → FFmpeg assembly with synchronized subtitles), all driven by code-defined SVG visuals (no stock or AI images). Exposes 4 automation levels (from stage-by-stage review to fully autonomous) and ships as a CLI with a dry-run cost estimator, plus a Dockerized build.

RoleRadar

Stack: Python, FastAPI, n8n, Streamlit, APScheduler, OpenAI (GPT-4o), BGE-M3 embeddings (sentence-transformers), Instructor, Pydantic v2, Docker Compose, Telegram

Autonomous job-matching radar that ingests raw postings from multiple job boards, normalizes and semantically deduplicates them, and honestly scores each one against my real profile—parsed directly from my CV PDF—via a hybrid heuristic fusing deterministic rules (role-ranking decay, seniority, location/remote, hard deal-breakers) with LLM judgment. Built around a single intelligent core shipped in two automation setups: (v1) the smart pieces run as a service while n8n handles the daily schedule, source ingestion, and digest delivery; (v2) a self-contained app that fetches, cleans, ranks, and pushes a Telegram/email digest on schedule with no human in the loop, behind a simple web UI. Scoring runs fully offline by default, with the LLM as an optional layer for written explanations—so it works end-to-end even without an API key.

FinFeed

Stack: Next.js, TypeScript, Python, Supabase, GitHub Actions, Groq (Llama 3.3 70B), OpenAI TTS, FFmpeg, faster-whisper

Autonomous multi-modal news pipeline that delivers AI-generated short-form videos twice daily across Finance and Tech categories. Engineered an end-to-end CI/CD pipeline (RSS ingestion → LLM script generation → TTS narration → animated subtitles → FFmpeg video synthesis → Supabase delivery) running in under 15 minutes with zero human intervention.

Florida Real Estate Price Prediction

Stack: Python, scikit-learn, XGBoost/LightGBM, Pandas, geospatial libraries

End-to-end ML pipeline for a Master's-level competition predicting Florida property prices on a dataset with ~60 features including JSON-nested attributes and geographic coordinates. Engineered domain-specific features (location clustering, property type encoding), benchmarked gradient boosting models, and optimized for MAE.