Skip to content

Projects

Ten deep-dives, one engineering standard.

Every card links to a full case study: the problem, the approach, the architecture, the numbers, and what I took away. All source is on GitHub.

Agentic AI2026

AEGIS Platform

An enterprise multi-agent system that behaves like an AI employee, not a chatbot.

Production-grade

  • Multi-agent
  • FastAPI
  • pgvector
  • Neo4j
  • Kafka
Read case study
LLM Infrastructure2026

OpenIncident

A production-grade MCP server suite that lets AI agents debug incidents without compromising security.

MIT licensed · 91 passing tests

  • MCP
  • OAuth2
  • OpenTelemetry
  • PostgreSQL
  • Kubernetes
Read case study
Medical AI · Reliability2026

Drift-Aware Medical AI

Detects and explains distribution shift in a chest X-ray classifier — with no new labels required.

Research / MLOps pipeline — not a clinical tool

  • PyTorch
  • Drift detection
  • FastAPI
  • Kafka
  • Calibration
Read case study
Applied ML · Fintech2026

CreditLens

A loan-decisioning platform that optimises expected profit per applicant, not AUC.

Production-grade reference implementation

  • LightGBM
  • CatBoost
  • Feast
  • MLflow
  • Fairlearn
Read case study
LLM Systems · Workflow2026

TRIAGE-FLOW

A self-hosted AI support-triage system built as production workflow engineering, not a demo.

Production-grade

  • n8n
  • TypeScript
  • PostgreSQL
  • Redis
  • Grafana
Read case study
LLMs · Fine-tuning2026

CLAUSE-CRAFT

A fine-tuned 7B model that extracts 41 contract-clause categories into validated JSON — with a full cost/accuracy analysis.

Benchmarked on real hardware

  • LoRA
  • SFT
  • DPO
  • vLLM
  • CUAD
Read case study
LLMs · Text-to-SQL2026

AskWarehouse

A text-to-SQL agent that self-corrects, asks for clarification, and refuses to guess.

Evaluated on 109 gold questions

  • Qwen2.5-Coder
  • DuckDB
  • dbt
  • sqlglot
  • Streamlit
Read case study
LLMs · RAG2026

FilingsIQ

Question answering over SEC 10-K filings that recomputes its own numbers and refuses to guess.

Evaluated on a 40-question benchmark

  • Qwen2.5-3B
  • BM25
  • bge-small
  • FastAPI
  • EDGAR
Read case study
Computer Vision · Video2026

Action Recognition with Vision Transformers

Video action classification and spatio-temporal localisation — with an honest negative result.

Advanced ML coursework — multi-seed results

  • VideoMAE
  • TimeSFormer
  • DETR
  • PyTorch
  • HMDB51
Read case study
NLP2026

Dialect-Robust Sentiment & Sarcasm Detection

Text classification that holds up across English varieties — where sarcasm inverts the label.

  • Transformers
  • NLP
  • Sarcasm
  • Robustness
Read case study

More work — including earlier RAG and fitness-tracking projects — lives on GitHub.