Mathematics & Computing @ IIIT Raichur
I build data-driven tools that extract insights, detect patterns, and automate decisions.
I'm a Mathematics and Computing student passionate about turning raw data into actionable insights. My work lives at the intersection of data engineering, machine learning, and intelligent web applications — I enjoy building systems that make sense of messy, real-world data.
- 🔭 Currently working on: Explainable & calibrated ML for healthcare risk prediction
- 🌟 Contributed a merged fix to matplotlib, backported to its stable release branch
- 🧮 Background in: Mathematical modeling and algorithmic thinking
- 🌱 Exploring: statistical rigor in ML (significance testing, subgroup fairness), NLP, and full-stack data apps
- 🤝 Open to: Collaborations on data/ML projects and open-source contributions
- ⚡ Fun fact: my first open-source PR review comment came from a matplotlib core maintainer
Languages
Machine Learning & Data Science
Frameworks & Libraries
Tools & Platforms
Rigorously validated ML pipeline for diabetes risk classification, with statistical significance testing and a subgroup fairness audit most published work in this space skips.
| Metric | Result |
|---|---|
| Dataset | 253,680 real patient records (CDC/BRFSS) |
| Best model (raw) | LightGBM, ROC-AUC 0.8254 (statistically significant edge over RF/XGBoost, bootstrap 95% CI) |
| Calibration improvement | Brier Score cut 43% (0.171 → 0.098) via Platt/isotonic scaling, with no loss in ranking ability |
| Explainability | SHAP values reproduced identically across independent runs (global + per-patient) |
| Fairness finding | Found and quantified real calibration gaps by Age (AUC 0.837 vs. 0.770) and Income |
| Engineering fix | Diagnosed and fixed a 441MB → 9.2MB model bloat bug via regularization |
Built with: Python scikit-learn XGBoost LightGBM SHAP Jupyter
| Project | What It Does |
|---|---|
| Text Insight Studio | Automatically profiles any uploaded CSV — detects column types, generates charts and word clouds, and produces natural-language summaries with no manual configuration |
| Suzume – Spotify Mood Analyzer | Detects a song's emotional mood in real time by combining live Spotify metadata with Genius lyrics sentiment analysis |
| Student Performance Risk Analyzer | Flags at-risk students using a fully interpretable, weighted-indicator scoring model — no black-box ML |
| Context-Aware AI Chatbot | Maintains full conversational memory across multi-turn sessions via a lightweight Python + JS architecture |
5 merged pull requests across open-source projects, including a fix merged into matplotlib — one of the most widely-used Python libraries in the world.
| Project | Contribution | Outcome |
|---|---|---|
| matplotlib (23.1k ⭐) | #31420 — Fixed a broken freetype dependency URL | Merged by a core maintainer; backported to the stable v3.10.x branch and shipped in milestone v3.10.9 |
| TermUI | #632 — Test suite for LogView, StatusIndicator, and Definition widgets | Merged, rated "quality: exceptional" by the maintainer (GSSoC 2026) |
| UltimateHealth | #1117 — Reusable glassmorphic Modal component | Merged — full ARIA support, focus trap, keyboard + backdrop close handling (GSSoC 2026) |
| statsmodels | #9766 — Improved docstrings in robust.norms (parameter descriptions, formatting, clarity per NumPy docstring standard) | Merged |
| Themix GUI (2.3k ⭐) | #562 — "Matcha Dark Sea" GTK color theme preset | Merged |
All contributions followed the standard open-source workflow: fork → feature branch → PR → maintainer review → merge.
I'm always open to interesting conversations about data, ML, and building things. Feel free to reach out!
- 📧 Email: [email protected]
- 💼 LinkedIn: manasvi-sahare
"Data is the new oil — but insight is the refined fuel."