CitySort
A runnable AI MVP to ingest city documents, extract fields, run rules/OpenAI/Anthropic classification, validate required fields, and route into department queues with human review.
- human review
- Anthropic
- audit trail
Data science + product engineering
Hey! My name is Imran. I build ML systems and products end-to-end.
MSc AI & Sustainable Development (University of Birmingham, Merit) and BSc Chemistry (Manchester Metropolitan, 2:1). Runner-up at the 2025 Cardiff NLP Hackathon.
- Speed
- prototype to production
- Quality
- tests, monitoring, polish
- Data
- analytics and ML systems
Currently building
- CitySort - rules engine + Anthropic automations
- OfflineAI - offline iOS LLMs (zero API cost)
- Fraud detection - production ML + monitoring
- Debate Chatbot - RAG over debate transcripts
GitHub: github.com/imranow
01
Skills
Tools and frameworks I use to build, ship, and monitor.
Programming & Data
- Python (NumPy, pandas)
- Git
- Data visualization (Matplotlib, Plotly, Streamlit)
- Vector databases: Pinecone
ML & Statistics
- Regression and classification
- Time series forecasting
- scikit-learn
- TensorFlow, PyTorch
NLP
- Transformers
- Zero-shot classification
- Sentiment analysis
- RAG (LangChain)
- LLM APIs (Anthropic Claude)
MLOps
- FastAPI
- Docker, Kubernetes
- CI/CD (GitHub Actions)
- AWS
- Monitoring (Prometheus, Grafana)
02
Projects
Short, scannable summaries with deeper case studies when you want them.
OfflineAI (Mobile App)
A privacy-first iOS app running small language models completely offline, with AES-256 encryption and intelligent resource management.
- offline LLM
- encryption
- resource mgmt
Fraud ML Detection
A fraud detection pipeline that turns raw events into ranked risk, with evaluation, monitoring, and feedback loops.
- ML systems
- evaluation
- monitoring
Debate Chatbot (RAG)
Grounded Q&A over 2019-2020 U.S. Democratic primary debate transcripts using Pinecone retrieval and Anthropic (Claude), returning citations in every response.
- RAG
- Pinecone
- Anthropic
03
Work experience
Highlights from recent roles. Full details are in the CV.
Cardiff NLP Hackathon
- Built and deployed a real-time NLP dashboard to analyse global trade news and identify impacted sectors for decision-making.
- Used zero-shot classification and sentiment scoring; shipped with Streamlit and interactive visuals (Plotly).
- Implemented a secure API key system and improved performance with caching and batch processing.
Elsevier
- Led data-driven organisational mapping of the Technology division to improve visibility and decision-making.
- Analysed engagement and onboarding feedback to improve internal processes.
- Presented insights to technical and non-technical audiences to support adoption and change.
Manchester Metropolitan University
- Consolidated 5 disparate IT databases into a unified knowledge management system.
- Supported system migration and data harmonisation under tight deadlines.
- Worked with stakeholders to diagnose inefficiencies and turn findings into process improvements.
04
Contact
If you want to collaborate or talk about any of these projects, reach out.