Work
9 projects
Client systems and personal work. Client projects are described generically — the architecture, decisions and trade-offs are real; the client is not identified.
Machine Learning & Data Science
2 projects lead with this
Feature engineering, supervised modelling, time series, NLP and vision classification, and the evaluation methodology that proves any of it worked.
- Role
- Sole author — time-series ML and MLOps
- Domain
- Machine Learning & Data Science
- Stack
- Python · LightGBM · ARIMA / ARIMAX
- Result
- ~4M order rows processed
- Role
- Sole author
- Domain
- Machine Learning & Data Science
- Stack
- Python · scikit-learn · pandas
- Result
- 85% held-out test accuracy
Generative AI & LLMs
3 projects lead with this
Prompt design, multimodal reasoning over text and images, structured output, output validation, and evaluating whether a generative system is actually correct.
- Role
- Owner — AI pipelines and all prompt engineering
- Domain
- Generative AI & LLMs
- Stack
- Python · FastAPI · Anthropic Claude
- Result
- ~75s livery composition pipeline
- Role
- Primary developer and top committer
- Domain
- Generative AI & LLMs
- Stack
- Python · Flask · RabbitMQ
- Result
- ~160s case summarization
- Role
- Sole developer
- Domain
- Generative AI & LLMs
- Stack
- Python · Flask · Google Gemini
- Result
- 343 learned query patterns
Data & Document Intelligence
2 projects lead with this
OCR, parsing, classification, entity and clause extraction, confidence gating, and validation — including deciding what a human still needs to see.
- Role
- Lead developer
- Domain
- Data & Document Intelligence
- Stack
- Python · FastAPI · AWS Textract
- Result
- ~93% routing accuracy
- Role
- Sole author — prompt, API, security and database
- Domain
- Data & Document Intelligence
- Stack
- Python · FastAPI · Anthropic Claude
- Result
- ~88% field extraction accuracy
Automation & Agents
2 projects lead with this
Workflow orchestration, event-driven triggers, multi-step agents, and the retry and idempotency logic that keeps them trustworthy.
- Role
- Built the entire system end to end
- Domain
- Automation & Agents
- Stack
- Python · Flask · Playwright
- Result
- ~25/day enquiries automated
- Role
- Sole builder
- Domain
- Automation & Agents
- Stack
- n8n · Retell AI · ElevenLabs
- Result
- ~30 automation workflows