Sashanka Deka

WORK HISTORY

EDUCATION

ABOUT SASHANKA DEKA

I spent 17 years building analytics systems that people actually used — forecasting models for 100K+ employee organisations, automation pipelines that saved man-hours a year, retail measurement science across LATAM markets. The through-line was always the same: turn messy data into decisions. In 2025 I took a deliberate sabbatical to go deep on Generative AI. Not courses for the sake of it — I built things. The first was a Document-Grounded Assistant — a production RAG system that answers questions strictly from your document, nothing made up. Sounds simple. Two months of my laptop overheating and a retrieval pipeline confidently fetching answers from completely the wrong pages taught me otherwise. The real problem in RAG isn\'t the LLM — it\'s getting the right context to it in the first place. I rebuilt the retrieval layer twice, eventually landing on PageIndex: the system identifies the right pages first, then fetches only those chunks. Corporate mode for training materials and policy docs, Academic mode for study material, both with page citations. Live on Streamlit, free, no signup. The other was an IPL 2026 Squad Analytics Dashboard — processing 8,952 ball-by-ball Cricsheet JSONs, live ESPN Cricinfo scraper, H2H matchup engine, percentile ranking within peer groups, team intelligence radar across 8 dimensions. Deployed on PythonAnywhere. Currently seeing visits a month and IPL hasn\'t started yet. The sabbatical wasn\'t a gap. It was a retool. I\'m looking for roles where data rigour and AI capability sit in the same room — analytics teams adopting GenAI, AI product companies that need someone who understands both the models and the business context, or organisations building something genuinely new with data at the core. MSc Statistics, University of Delhi. 17 years across IEG, Ameriprise, TCS/Nielsen, Cognizant, HCL. Now building in public. sashankadeka.pythonanywhere.com 🤖 document-grounded-assistant.streamlit.app substack.com/@sashankadeka