Ignacio Ordovás Pascual

Data science portfolio

Work

Case studies from consulting, work that was made public, and things I've built on my own to learn.

Case studies

Some of the client work I've participated in.

Pharmaceutical company · 2025–2026

Grounded content generation for pharma

Problem
Teams needed emails, slide decks and medical documents that stay faithful to approved material, plus a research assistant over internal knowledge.
Approach
A research-assistant chatbot and a content generator grounded on approved sources, together with a second solution that reviews the generated content.
Outcome
User satisfaction around 40% higher than with the previous tools.
  • GenAI
  • NLP
  • Deployment

Banking · 2024

RAG assistant over internal knowledge

Problem
Employees had to search scattered internal documentation to answer day-to-day questions.
Approach
A retrieval-augmented chatbot over internal data, iterated in production to improve answer quality.
Outcome
Adoption grew from 50 to 600 employees.
  • GenAI
  • NLP
  • Deployment

Logistics planning · 2023-2024

Conversational analytics for supply chain

Problem
Planners needed quick descriptive analytics over supply-chain data without writing queries.
Approach
A generative-AI chatbot connected to a Neo4j graph database that answers questions with automatic descriptive analytics and plots (PoC).
Outcome
Proof of concept developed ready for further testing and deployment and to be presented to clients in logistics and supply-chain management.
  • GenAI

Energy · 2021

Rust detection for asset maintenance

Problem
Experts reviewed drone footage of petrol tanks by hand: a one-day inspection could take three to four days to report.
Approach
A U-Net segmentation model that finds rust down to pixel level, served through an API that analyses the uploaded drone video.
Outcome
About 95% accuracy, turning days of manual review into near-real-time reports, as part of a solution aiming to cut inspection costs by 50%.
  • Computer vision

Blog post

Published work

Work that was made public by clients, employers or partners.

Article + code · 2026

Injecting company jargon for generative AI

How to inject company-specific vocabulary from a YAML file into LLM prompts, using an app that turns natural-language requests into arXiv API queries, and how well it works across model sizes.

  • GenAI
  • NLP

Summary Code

Case study · 2022

Retail self-checkout with Azure Percept

A general fruit-detection model on Microsoft's Azure Percept edge-AI device, using Custom Vision and IoT Hub, built during a Microsoft bootcamp.

  • Computer vision
  • Deployment

Blog post Video

Personal projects

Things I built on my own to learn.

Project

Japanese video-game sales

Web scraping and regular expressions to mine Japanese video-game hardware and software sales.

  • Data Science

Code

Project

Amazon book ratings from review text

Predicting Amazon book ratings from the review text alone.

  • NLP
  • Data Science

Code

Learning archive

Read the project descriptions

Projects from the Ironhack Data Analytics bootcamp (Oct–Dec 2020), where I moved from astrophysics into data science.

  • Dice score recognition in images and live videoFinal project, selected for the HackshowCode
  • Kaggle competition: predicting diamond pricesCode
  • Chat API and text sentiment analysisCode
  • Geospatial analysis: choosing the best location for a gaming companyCode
  • The 500 best albums of all time, with the Spotify APICode
  • A word-frequency analysis of shark attacksCode
  • Hangman in Python (moddable)Code

More exercises, learning notebooks and experiments are on my GitHub profile.