Data Engineer · MSc Computer Science (University of Bath) · ML Robustness Researcher
I'm a data engineer with 2+ years of industry experience at Accenture and a Master's in Computer Science from the University of Bath, where my dissertation studied why deep video models break — and how to measure it. I build reliable data platforms by day and study robust, interpretable machine learning by night.
🔭 Currently: open to Data / ML Engineering roles in the UK and PhD opportunities (UK · Germany · Europe) in robust & interpretable deep learning for video understanding.
My MSc dissertation (supervised by Dr Davide Moltisanti, University of Bath) evaluated the robustness of I3D action-recognition models under targeted vs. random video perturbations, guided by Grad-CAM saliency.
Key result: perturbations localised to Grad-CAM-salient regions collapsed prediction consistency to 0% (vs. 60% for equivalent random perturbations) — evidence that I3D leans on a small set of fragile, motion-sensitive regions.
📄 Code, full results and the dissertation PDF: action-recognition-robustness
Research interests: robustness & evaluation of video understanding models · model interpretability (Grad-CAM, attention analysis) · failure analysis under distribution shift · reproducible ML research tooling
| Project | What it is | Stack |
|---|---|---|
| action-recognition-robustness | MSc dissertation: Grad-CAM-guided perturbation testing of I3D on UCF101 | PyTorch · I3D · Grad-CAM · OpenCV |
| videorobust | Open-source toolkit for robustness evaluation of video classifiers — perturbations, metrics & experiment harness (grown out of my dissertation) | Python · PyTorch · pytest |
| football | End-to-end batch ETL: 10 EPL seasons → star-schema warehouse, orchestrated & quality-gated | Airflow · PySpark · Postgres · Docker · Power BI |
| retail-insights-platform | Modern ELT lakehouse: ingestion → dbt marts → tested, documented analytics models | dbt · DuckDB · Python · pytest · GitHub Actions |
| sociolex | Lightweight NLP toolkit for extracting & clustering psychosocial language (CLI + FastAPI) | Python · scikit-learn · FastAPI |
Data Engineering — Airflow · Spark/PySpark · dbt · SQL (Postgres, DuckDB) · ETL/ELT design · star-schema modelling · data quality & testing
ML / Research — PyTorch · scikit-learn · computer vision (video models, Grad-CAM) · perturbation & robustness evaluation · experiment design
Engineering — Python · Docker · GitHub Actions (CI) · pytest · FastAPI · Linux
- 💼 LinkedIn: odduri-vishwaksha
- 🎓 Dissertation: PDF
- 📫 Reach me via LinkedIn or a GitHub issue on any of my repos
If you're a recruiter or a prospective PhD supervisor — the two pinned repos below are the best place to start.