Available for Internship · B.Tech CSE (AI & ML) · Manav Rachna · Batch 2027
KRISHNABHATIA
Data Science & AI Engineer · Building Explainable, Deployed ML Systems
I build ML systems that go beyond accuracy — they explain why.
My AQI Early Warning System is live in production, predicting air quality for 26 Indian cities using XGBoost + SHAP, serving real-time advisories for 7 vulnerable population groups. Not a notebook. A product.
✓ Oracle OCI Gen AI Professional 2025✓ Oracle OCI Data Science Professional 2025✓ Microsoft Azure AI-900
// 01 · FLAGSHIP PROJECT
THE BUILD THAT MATTERS
🔴 LIVE · DEPLOYED · BUILD V2 · PWA
AI-DRIVEN EARLY WARNING SYSTEM FOR URBAN AQI
With Explainable AI · 26 Indian Cities · XGBoost R²=0.932
A full-stack ML platform that predicts Air Quality Index for 26 Indian cities in real-time, explains every single prediction using SHAP game-theory explainability, and delivers personalised health advisories for 7 vulnerable population groups — children, elderly, asthma/heart patients, pregnant women, athletes. Served through a production FastAPI backend, deployed globally on Vercel. Free forever.
What no existing platform has — CPCB, SAFAR, IQAir, Breezometer — is an open, explainable, ML-driven system. This project fills that gap and is now live.
XGBoost 2.0SHAP TreeExplainerFastAPI + UvicornWAQI Live APIGoogle Gemini AIReact + VercelPWAAQI Time MachineMLflow TrackingFolium + PlotlyPython 3.13
MODEL PERFORMANCE
AlgorithmXGBoost
R² Score0.932
MAE17.928
RMSE29.500
5-Fold CV R²0.931 ±0.003
Training Records24,850
vs Random Forest+0.003 R²
SHAP FEATURE IMPORTANCE
PM2.50.4686
CO0.3059
PM100.0641
Year (trend)0.0389
26
Cities Live
24,850
Training Records
0.932
R² Score
7
Health Risk Groups
V2
Build · PWA
🧠
Explainable AI
Every prediction decomposed by SHAP game-theory. "PM2.5=180 → AQI +152.7 pts. CO=12.9 → AQI +82.9 pts." No other public platform does this. Recruiters see the WHY, not just the WHAT.
⚡
Live PWA + Time Machine
Progressive Web App with 26 live WAQI stations, threat matrix ranking all cities by AQI, city comparison tool (e.g. Mumbai 88% worse than Delhi), and AQI Time Machine for historical pattern analysis.
🏥
Health Intelligence
Personalised risk matrix for children, elderly, asthma/heart patients, pregnant women, athletes — across all 6 CPCB categories (Good→Severe). Precautions generated per reading. Nothing in market has this.
⚙️
Production Backend
FastAPI REST API with 3 production endpoints. Pydantic v2 validation, auto-Swagger docs at /docs. XGBoost + SHAP served as pickle artifacts. Deployed on Render.com. MLflow experiment tracking throughout.
🤖
AI Chatbot
Google Gemini 1.5 Flash integration for natural language AQI queries. "Is it safe to go running in Jaipur today?" → AI-generated personalised response using live AQI + health profile context.
📊
Manual Predictor
Any user can enter custom sensor readings (PM2.5, PM10, CO, NO2, NOx, SO2, O3, NH3) and get an instant ML prediction + SHAP explanation + health advisory. Works for cities CPCB doesn't cover.
// 02 · OTHER PROJECTS
MORE BUILDS
🛸
TRINETRA
AI-powered space exploration assistant. Integrated NASA REST APIs to fetch and visualize planetary data for all 8 planets via automated Python data pipelines and a Streamlit dashboard with ML predictions.
NASA APIsStreamlitPython PipelinesOngoing
🫁
MEDICAL CNN
TensorFlow CNN on 7,000 chest X-rays achieving 95% validation accuracy across COVID, Pneumonia, and Normal. Data augmentation and class imbalance handling for clinical reliability.
95%
Validation Accuracy · 7K X-rays
TensorFlowCNNOpenCV
❤️
HEART FAILURE
Benchmarked 7 ML algorithms on 918 patient records with 11 clinical features. XGBoost optimized with F1-Score + AUC-ROC ensuring clinical reliability — recall for heart failure cases prioritized over raw accuracy.
Coordinated placement drives for 60+ students, directly liaising with companies to streamline end-to-end campus recruitment. Managed company communications, scheduling, and student-side preparation — developing negotiation, stakeholder management, and communication skills in a high-stakes environment.
June 2025 – July 2025
AI & AUTOMATION INTERN
Unsaid Talks
Delivered 4 production-ready AI automation pipelines across 3 frameworks — LangChain, Agno, and n8n — reducing manual effort for study planning, content creation, and recipe generation workflows to near zero.
January 2025 – Present
ML & DESIGN HEAD
MRAC AIML Club · Manav Rachna IIRS
Led the club's technical activities, organising a Kaggle ML competition for 30+ students building a Spotify recommendation system — the highest-engagement session of the year. Mentors juniors through guided, project-based ML learning rather than pure tutorial consumption.
Special Mention · IIIT Delhi Hackathon 2024AI-based misinformation detection system recognised at national-level hackathon competing against hundreds of teams across India.
🥇
1st Position · College Mathematical FestPoster "Mathematics in Music" — exploring mathematical structures underlying music theory, rhythm, and composition.
📜
Certificate of Appreciation · Haryana State Productivity CouncilConcept paper "Intellectual Property Protection Strategy for Startups" recognised at state government level.