AI & Machine Learning Engineer | MSc Artificial Intelligence @ University of Manchester
Building deep learning systems that outperform benchmarks and deploy in the real world.
I'm an AI/ML engineer with hands-on experience across deep learning, RAG, LLMs, and reinforcement learning. Currently completing my MSc in Artificial Intelligence at the University of Manchester, I focus on designing models that deliver in the real world by solving existing problems.
- Deployed a Safe RL navigation system tested at TATA Motors India — published in ScienceDirect
- Built a 3D point cloud classifier achieving 94.4% on ModelNet10 — outperforming state-of-the-art by 1.3% — published in IEEE Xplore
- Programmed a YuMi robot to solve a Rubik's Cube at ABB India using optimised CV + solving algorithms
- Aerial image segmentation via hybrid Swin-UNet — Springer Nature Link
Languages: Python · C/C++ · SQL
Deep Learning: PyTorch · TensorFlow · Keras · Hugging Face Transformers
NLP / LLMs: Attention Mechanisms · BiLSTM · Transformer Architectures · Sentiment Analysis
Reinforcement Learning: TD3 · PPO · Deep Q-Networks · ROS2 · Gazebo
Deployment & MLOps: Docker · Streamlit · FastAPI · TensorBoard
Tools: ROS · RobotStudio · Jupyter · MATLAB · Git
| Project | Description | Stack |
|---|---|---|
| Swin-UNet Semantic Segmentation | Hybrid Swin Transformer + UNet for aerial image segmentation. Outperforms SegNet and DeepLabV3+ on the MBRSC Dubai satellite dataset. | PyTorch · Keras · Swin Transformer |
| Attention-Augmented BiLSTM Sentiment Analysis | NLP sentiment classifier combining a BiLSTM backbone with multi-head attention for improved long-range context modelling. | Keras · TensorFlow |
| Alpha Pipeline | End-to-end quant data pipeline: OHLCV ingestion into DuckDB, RSI/momentum/volatility feature engineering, vectorised SMA-crossover backtesting, and an interactive Streamlit dashboard. Dockerised with CI/CD and ~70 unit tests. | Python · DuckDB · Streamlit · Docker |
| Market Anomaly Agent | Autonomous LangGraph ReAct agent that detects statistical anomalies in equity data (Z-score, CUSUM, Isolation Forest) and investigates them using news sentiment, macro context, and quant signals. Deployed on AWS Lambda with EventBridge scheduling. | LangGraph · FastAPI · AWS · Python |
| RAG Culinary Assistant | Production-grade RAG pipeline for East Asian cuisine Q&A, built from scratch without LangChain or LlamaIndex. Hybrid FAISS + BM25 retrieval fused via RRF, cross-encoder reranking, semantic chunking, and 23 evaluation metrics across 7 ablation configurations. | PyTorch · FAISS · Gradio · HuggingFace |
| UK Legal Assistant | RAG system answering UK legal questions in plain English, grounded in GOV.UK and Citizens Advice documents. Hybrid FAISS + BM25 retrieval with Llama 3.1 8B, FastAPI backend, and a pure-CSS frontend. Deployed on Railway. | Llama 3.1 · FAISS · FastAPI · Python |
| Paper | Venue | Year |
|---|---|---|
| A Comprehensive Review on Safe Reinforcement Learning for Autonomous Vehicle Control in Dynamic Environments | ScienceDirect | 2024 |
| Point Cloud-Based 3D Object Classification With Non Local Attention and Lightweight Convolution Neural Networks | IEEE Xplore | 2024 |
| Enhanced aerial image segmentation via hybrid Swin-UNet with dilated convolutions and multi-scale fusion | Springer Nature Link | 2025 |
Open to ML Engineer, Data Scientist, and AI/LLM roles — UK & remote.