B.Tech Data Science & Computer Science, VIT Vellore (2026).
I build data and ML systems, and I'm more interested in whether a result holds up than in whether it looks good. Most of the repositories below contain at least one measurement that came out the wrong way and stayed in the README.
| Project | What it is |
|---|---|
| rxinsight | A pharmaceutical commercial analytics warehouse. PostgreSQL star schema, Slowly Changing Dimensions, a streaming Python ETL over 2.2M rows, and window-function SQL. 36 tests. |
| MedGraph | Agentic clinical document intelligence. A LangGraph state machine that answers from public healthcare policy and drug-label corpora — or refuses, with a stated reason, when the corpus can't support an answer. |
| anchor | A retrieval-augmented QA and agent service whose centrepiece is the measurement layer that says where its retrieval actually fails. Runs with no API keys and no network calls. |
| modelwatch | When a drift monitor fires, has the model actually degraded? Benchmarks drift detectors against realised degradation on real US Census windows, with ground truth the detectors never see. |
| mlops-portfolio | Model serving end to end — FastAPI behind Docker, real-time inference over Kafka, Terraform for infrastructure, deployed to Kubernetes. |
| visiondoc-ai | Vision-language document understanding, LoRA fine-tuned on consumer hardware and benchmarked against a zero-shot baseline. |
Languages Python, SQL Data PostgreSQL, pandas, PySpark, Databricks, ETL and dimensional modelling Infrastructure Docker, Kubernetes, Terraform, AWS, FastAPI, GitHub Actions ML PyTorch, TensorFlow, scikit-learn, LoRA/PEFT, RAG, LangGraph
Finishing my B.Tech and looking for data engineering and ML engineering work.