supply_integration
Production ELT for 100+ heterogeneous supplier feeds.
- Python
- SQL
- Apache Airflow
- dbt
- BigQuery
- GCP
Work
Selected projects
Selected projects, in rough order of how much they say about how I work. Each one has its own page: what it is, why it needed building, how it works, and what I would change.
catalog_data_platform
An ELT pipeline for promotional-goods catalogue data: supplier feeds to S3 to dbt/DuckDB to Streamlit.
- Python
- Apache Airflow
- Astronomer
- dbt
- DuckDB
- AWS S3
- Streamlit
exam-nation
Misconception-driven Socratic tutoring, one codebase across six courses and two exam systems.
- Next.js 15
- TypeScript
- Supabase
- Anthropic
- Vitest
- Stripe
- Vercel
IELTS Writing Task 1
AI-graded IELTS Academic Task 1 preparation, organised into hand-built lessons.
- Next.js 16
- Supabase
- OpenRouter
- Stripe
- Vercel
CEFR speech scoring lab
ASR, fluency features, agreement and fairness evaluation for L2 speaking assessment.
- Python
- Whisper
- scikit-learn
- PyTorch
- MLX
- praat-parselmouth
- jiwer
PDP parity validator
Seven-dimension regression detection across a site replatform, with LLM root-cause hypotheses.
- Python
- Streamlit
- Playwright
- Anthropic
- OpenRouter
Portfolio evidence RAG
Local hybrid retrieval over my own repositories, built to measure what hybrid search and reranking actually buy.
- Python
- sentence-transformers
- SQLite
- sqlite-vec
- pgvector
- Anthropic
- Langfuse
Student request triage
An LLM pipeline that classifies and answers student support requests, built to stay correct when the services behind it misbehave.
- Python
- httpx
- Pydantic
- SQLite
- pytest
- respx
- Docker Compose