Data & Analytics Engineer. I build robust data pipelines, read source code, and ship fixes upstream.

MSc Data Analytics. Building pipelines and dev tools on the side. I believe compliance shouldn't mean spreadsheets and AI shouldn't require the cloud. Yorkshire, UK.

Portfolio

OpsMind — On-prem AI query tool for manufacturing. docs

• Ask production questions in plain English, get SQL results in 5 seconds
• LangGraph multi-step agent (6-node state graph) with 5-stage SQL validation
• MCP server architecture: database + doc search as decoupled tool servers
• pgvector + ChromaDB retrieval, runtime-loaded domain docs
• Gemma 3 12B via Ollama — no data leaves the factory
• 7 business domains, formal agent specs, ty type checker in CI
• Docker deployment with isolated Ollama container, structured JSONL audit logging
• Golden-set eval harness (library + LLM paths) with failure-mode taxonomy
• Governance, security policy, and code of conduct published — first-PR-wins assignment

Production Analytics Pipeline — Incremental ETL from fish production ERP

• 15K+ rows daily from 4 ERP tables, validated with Pydantic
• FastAPI REST API (11 endpoints) + Next.js dashboard + Power BI export
• Prefect orchestration, Sentry monitoring, Docker + OpenTofu deployment
• Batch tracking, yield analysis, shelf life management, traceability | 53 tests
• Apache 2.0 licensed; governance, security, and code-of-conduct documents published

UK Crime Pipeline — Police UK API to PostgreSQL and BigQuery. streamlit / looker studio / hugging face

• 99,675 records, 10 cities, 6 dbt marts (including outcome analysis and YoY trends), 65 tests
• Declarative data validation + SLO monitoring (freshness, completeness, volume)
• Polars-based alternative ingestion, pipeline maturity scorecard
• 3 CI/CD workflows with ty type checker, diskcache + stamina for API resilience
• Apache 2.0 licensed; NOTICE documents the OGL-v3.0 chain on derived datasets

sql-sop — SQL linter on PyPI. pip install sql-sop

• 23 rules (10 errors, 13 warnings) covering DELETE/UPDATE-without-WHERE, implicit cross joins, nested subqueries, unused CTEs, SELECT *, and more
• 78 tests, sqlparse AST parsing, fluent API (SqlGuard().enable(...).scan(...))
• libCST-based Python scanner catches SQL injection in .execute() / .read_sql() calls (v0.4.0)
• Pre-commit hook + GitHub Action for CI/CD integration, 195+ monthly downloads
• MIT licensed (deliberately kept — PyPI downstream stability); full governance + security policy published

I learn tools by reading their source. I reverse-engineered the drt connector architecture, shipped 5 destination connectors, and wrote the official connector tutorial — all merged. Same approach everywhere: read the internals, find the gap, ship the fix.

drt · pandas · ChromaDB · pgcli · ollama · superset · plotly · fpdf2

Python, SQL, dbt, PostgreSQL, BigQuery, FastAPI, Streamlit, Prefect, LangGraph, Ollama, Docker, Polars, pandas, Pydantic, pytest, GitHub Actions