Ben DruryIn development
SpecLogic
Data-first buying decisions
A decision engine for high-ticket purchases, turning complex specs into comparable scores so buyers choose with confidence, not guesswork.
- Comparison engine
- Technical benchmarks
- Category scoring
Senior Data Scientist, AI & Automation
Ben Drury
I turn messy, real-world data problems into production systems, from scoping and modelling through to deployment. I measure success by impact delivered, not complexity built.
At Cromwell Tools on AI extraction and structured content. Previously led automation and ML at Zoro UK.
Current role
Data Scientist
Cromwell Tools
How I work
Production ML
Scope, ship, measure impact
Education
MSc Data Analytics
Distinction, De Montfort University
Portfolio
Interactive projects & demos
Live implementations you can explore: RAG, LLMs, multimodal validation, and more.
RAG Age Restriction Identifier
Hybrid retrieval across UK legislation with confidence scoring and evidence citations.
RAGFastAPIGCP
LLM Product Attribute Extractor
Structured attribute extraction from messy product content with hybrid targeting modes.
LLMPythonDocker
Multimodal Product Validator
Embeddings-first validation with Gemini fallback in the gray zone.
EmbeddingsVertex AIGCP
RAG Document Q&A
Upload PDFs and chat with LangChain + FAISS retrieval and Firestore history.
RAGLangChainFirestore
Beyond this site
Building & live work
Separate from portfolio demos: products and platforms I'm shipping elsewhere, including work in progress.
Architecture & delivery
Data science work without a demo
Production systems where the value is in pipelines, scale, and outcomes, not a clickable UI on this site.
Zoro UK · ML
Hybrid Classification Pipeline
Auto-categorised 5M+ records across 2,000+ categories at 84% accuracy, halving manual categorisation and accelerating supplier onboarding.
Product data
Classifier
Taxonomy
- 84% accuracy at scale
- 50% less manual work
- Faster supplier onboarding
PythonNLPBigQuerySQL
Zoro UK · Entity resolution
Vector Product Grouping
Vector-based model grouped 70% of SKUs across 300K+ families at 90%+ accuracy, reclaiming 30+ hours of manual effort per week.
SKU embeddings
Similarity
Product families
- 90%+ grouping accuracy
- 70% SKUs auto-grouped
- 30+ hrs/week saved
PythonVector searchEmbeddingsGCP
Cromwell Tools · Current
Structured Content & Extraction
AI extraction from PDFs and supplier feeds, plus a componentized content framework replacing unstructured blobs with reusable fields.
PDFs & feeds
Extract + normalise
Structured PIM
- Automated PDF/supplier extraction
- Structured page fields
- Cross-channel consistency
PythonLLMsETLGCP
Career
Experience & education
A quick scroll through roles and study. Full detail in the CV.
- workDec 2025 to Present
Data Scientist
Cromwell Tools
Building AI extraction from PDFs and supplier data, leading a structured content framework for product pages, and normalisation models for attribute consistency across channels.
- workOct 2023 to Dec 2025
Senior Data Automation Analyst
Zoro UK (Grainger plc)
Technical ownership of automation for 5M+ products: hybrid classification, vector-based product grouping, mentoring, and presenting ROI to global leadership.
- workNov 2022 to Oct 2023
Data Automation Analyst
Zoro UK (Grainger plc)
Python ETLs and supplier pipelines; developed a TF-IDF alternative-product model, A/B tested via Unleash, and scaled to full site traffic after significance.
- workMar 2022 to Nov 2022
Content Quality Analyst
Zoro UK (Grainger plc)
Cut handover-to-launch to 3 days with onboarding automation (1M+ image downloads) and Looker/BigQuery dashboards for catalogue quality.
- educationOct 2020 to Feb 2022
MSc Data Analytics (Distinction)
De Montfort University, Leicester
Advanced analytics, statistics, and data systems with applied projects.
- educationOct 2017 to Jul 2020
BSc (Hons) Computer Science (First Class)
De Montfort University, Leicester
Software engineering and data foundations with practical, project-led focus.
Toolkit
Tech stack & strengths
Tools and domains I work in day to day, from notebooks to production APIs.
Python
Languages
SQL / BigQuery
Data
LLMs & RAG
AI
FastAPI
Backend
GCP
Cloud
Docker
DevOps
ETL Pipelines
Data
Vector Search
ML
Git / CI
Engineering
Vertex AI
ML
Systems
Python · SQL · FastAPI · Docker · GCP (Cloud Run, Vertex AI, BigQuery)
ML & AI
Vector search · NLP · Clustering · Information retrieval · Multi-modal extraction
Delivery
Technical roadmapping · Stakeholder management · Agile delivery · Production monitoring
Certifications
Databricks Gen AI & AI Agents · Google Cloud Generative AI · TensorFlow