My Projects

My top 6 projects: risk analytics, forecasting, multi-agent AI, NLP, and GenAI/RAG systems.

Risk and Advanced Modelling pharmacovigilance dashboard Risk Analytics

Risk & Advanced Modelling

Four independent risk-analytics modules for a regulated setting, each pulling real, live, public regulatory data (openFDA, ClinicalTrials.gov, US Treasury OFAC), built on a shared risk-scoring/ML framework, and surfaced through an interactive dashboard with stakeholder-facing reports.

  • Tech Stack: Python, Streamlit, scikit-learn, RandomForest, Logistic Regression
  • Focus: Pharmacovigilance signal detection, recall risk, clinical trial risk, and sanctions screening
  • Outcome: Four end-to-end notebooks run against live data, with a shared risk-scoring engine and a stakeholder dashboard.
View on GitHub
Forecasting

Retail Demand Forecasting

A production-grade demand forecasting system built for Tesco-scale retail operations, demonstrating deep learning forecasting models, robust backtesting pipelines, probabilistic forecasting, and hierarchical reconciliation across 50 stores, 200 products, and 3 years of daily sales data.

  • Tech Stack: Python, PyTorch, PySpark, Django, Docker
  • Focus: LSTM/GRU, TCN, Transformer (TFT-inspired), N-BEATS, and DeepAR forecasting models, plus MinTrace reconciliation
  • Outcome: A backtested, production-style forecasting pipeline with an interactive Django dashboard.
View on GitHub
BBC News NLP Intelligence Platform showing classification, entity extraction, and similar-article results NLP

BBC News NLP Intelligence Platform

A fully local NLP platform for the BBC News archive: dataset auditing, fine-tuned topic classification, zero-shot sub-topic tagging, named entity recognition, and semantic search, served through both an API and a web interface with no external API calls.

  • Tech Stack: Python, ModernBERT, GLiNER, Jina Embeddings, FastAPI, Django, Docker, MLflow
  • Focus: Fine-tuned transformer classification, zero-shot sub-topic assignment, entity extraction, and a local semantic search index
  • Outcome: 89.84% classification accuracy at ~154ms latency, backed by unit/integration/evaluation tests, Dockerized deployment, and MLflow experiment tracking.
View on GitHub
Insurance AI Service Swagger API docs from the Building Agents portfolio AI Engineering

Building Agents Portfolio

Production-oriented multi-agent AI systems across regulated domains: insurance claims triage & SEC filing research, healthcare prior-auth extraction & CDC BRFSS research, and fintech AML alert scoring & CFPB complaint research.

  • Tech Stack: Python, FastAPI, LLM Agents, Docker, CI/CD
  • Focus: Multi-agent orchestration combining LLM reasoning with deterministic fallback logic and external data tools
  • Outcome: Three domain-specific agent systems built with production-style engineering gates (CI, containerisation, HTTP APIs).
View on GitHub
GenAI / RAG

WHO RAG Service

A microservice for querying WHO guideline documents using Retrieval-Augmented Generation: PDF ingestion, SBERT embeddings, vector indexing, and semantic summarisation via an HTTP API.

  • Tech Stack: Python, FastAPI, FAISS, Pinecone, SBERT
  • Focus: Document ingestion & chunking, embedding generation, vector search, and RAG-based summarisation
  • Outcome: Shipped an /ask endpoint for semantic search and summarisation over WHO guidelines, with pytest coverage.
View on GitHub
Health AI / Agents

MedBot — Symptom Triage Assistant

A medical symptom-triage assistant that layers a rule engine, an ML classifier, and an LLM fallback to assess urgency and provide contextual guidance, with SHAP-based explainability dashboards so every decision stays auditable.

  • Tech Stack: Python, FastAPI, scikit-learn, spaCy, HuggingFace Transformers, SHAP, Docker
  • Focus: NLP symptom extraction, rule-based risk flags, ML severity classification, and transparent LLM-backed guidance
  • Outcome: A CI-tested, Dockerized triage API (pytest + GitHub Actions) with SHAP explainability and OpenAPI docs, released under MIT license.
View on GitHub