Assistant Professor, AI & ML at SNMIMT, Ernakulam
I work on AI, machine learning, data science, computer vision and analytics, and on the research and teaching around them. My projects are independent studies on public data, and interactive tools I build for my students.
Numbers carry their status: Verified means a committed, dated run in the repo; Reported means taken from a paper and not reproduced.
- Anomaly detection for monitoring: on injected anomalies, false alerts fall 77% against a z-score at matched recall; on real expert-labelled data no detector is distinguishable from it (Verified, run 2026-10-14). Code · Live demo
- E-commerce recommender: EASE reaches hit rate@10 of 46.1% on Online Retail II, +60.2% over popularity (Verified, offline, run 2026-10-14). Code · Live demo
- Transfer learning for vision: MobileNetV2 reaches the target accuracy in 60% to 76% less CPU training time than a scratch CNN (Verified, two committed runs, 2026-10-14 to 17). Code · Live demo
- Pneumonia X-ray CNN: test AUC 0.986 for a small CNN against 0.928 for a pixel logistic-regression baseline on PneumoniaMNIST; a research exercise, not a diagnostic tool (Verified, 5 seeds, run 2026-10-15). Code · Live demo
- RAG document QA evaluation: hybrid retrieval with reranking reaches Recall@5 of 0.973 on held-out SQuAD 2.0 questions (Verified, run 2026-10-15). Code · Live demo
- Loan credit-risk prediction: on German Credit no model beats logistic regression reliably (AUC 0.798); the cost-based threshold costs 0.520 per applicant against 0.700 for rejecting everyone (Verified, run 2026-10-15). Code · Live demo
- Analytics dashboards: retail, HR and marketing dashboards on public data; 33 of 33 data-quality checks pass (Verified, run 2026-10-15). Code · Live demo
- Black pepper recognition: an explainer of my co-authored IEEE paper, which reports 100% global accuracy on 115 test images (Reported in the paper, not reproduced here). Code · Live demo
SMS spam (NLP) · Face-mask detection · FX and Treasury forecasting (LSTM, DQN) · Model serving with FastAPI · Housing price regression
NetVerse · dsa · bit · ML course notebooks · PDF to Quiz · Fake-News · AIIDS
| Area | Tools and methods |
|---|---|
| Generative AI & LLMs | LLM APIs (Gemini, OpenAI, Anthropic), prompt engineering, retrieval-augmented generation, embeddings and reranking, Hugging Face Transformers, structured outputs |
| Machine & deep learning | Supervised and unsupervised learning, deep learning, NLP, time-series forecasting, anomaly detection, recommendation systems, reinforcement learning, explainable AI (SHAP), Optuna, scikit-learn, XGBoost, PyTorch, TensorFlow, Keras |
| Computer vision | Image processing, face detection (MTCNN), transfer learning, CNNs, OpenCV |
| Data science & BI | Pandas, NumPy, SciPy, Matplotlib, Seaborn, Plotly, EDA, statistical analysis, data visualisation |
| Data engineering | SQL, ETL / ELT pipelines, data modelling, pipeline automation |
| MLOps & cloud | MLflow, Docker, Git and GitHub, CI/CD with GitHub Actions, FastAPI, Flask, REST APIs, model serving and monitoring |
| Programming | Python, SQL, C / C++, MATLAB, JavaScript / TypeScript, React, Next.js, Pydantic, Pytest, Linux, software testing |
| Research | Experimental design, statistical evaluation, benchmarking, ablation studies, reproducible experiments, literature review and research-gap analysis |
| Web & visualisation | React, Next.js, Three.js, interactive data visualisation, 3D visualisation, interactive simulations |
| Familiar with | LangChain, LlamaIndex, vector databases, LLM fine-tuning (LoRA / PEFT), AI agents, LLM evaluation, LIME, YOLO, U-Net / SAM, ViT / CLIP, ONNX Runtime, TensorFlow Lite, Dash, Streamlit, DVC, Power BI, Tableau, Microsoft Fabric, Databricks, PySpark, Apache Spark, Airflow, MySQL, PostgreSQL, MongoDB, Azure Data Lake, AWS, Azure, Kubernetes, Java, Weights & Biases |