3Kernels

• 

  Architecture Diagram

Inspiration Millions of Indian railway passengers face delays daily but have no idea they're entitled to free meals, refunds, or alternative transport under Railway General Rules. We wanted to bridge this gap — making rights accessible to every passenger, in their own language. What it does Yatri Sahayak predicts train delays using historical data and a GBT Regression model, then automatically maps the predicted delay to the passenger's legal entitlements. Results are displayed in English, Hindi, and Tamil through an interactive dashboard — all running on Databricks. How we built it We ingested Indian Railways delay data into Delta Lake, engineered features using PySpark, and trained a GBTRegressor with Spark MLlib. A rules engine stored as a Delta table maps delay buckets to Railway General Rules. MLflow tracks experiments. The front-end is a Databricks notebook with widgets and a custom HTML dashboard rendered via displayHTML. Challenges we ran into The initial dataset had no delay column in a usable format — we had to pivot our approach mid-hackathon. MLflow model logging failed on serverless clusters due to UC volume requirements, costing us debugging time. Scoping down from real-time streaming to batch prediction was a tough but necessary call under time pressure. Accomplishments that we're proud of We built a fully working end-to-end pipeline — from raw CSV to multilingual rights output — in under 2.5 hours. The dashboard looks polished, the model runs, and the rules engine actually maps to real Railway General Rules. No mock-ups, everything runs live. What we learned Scope ruthlessly. We planned for RAG pipelines and real-time APIs but shipped a clean batch pipeline instead. We also learned how deeply integrated Delta Lake, MLlib, and MLflow are within Databricks — the platform made the tight timeline possible. What's next for 3Kernels Integrating real-time delay data via NTES/Indian Rail API, adding Param-1 or IndicTrans2 for dynamic multilingual generation across 22+ Indian languages, deploying as a standalone Databricks App, and expanding coverage to metro and bus networks.

Built With

• deltatable
• genie
• mlflow
• mllib
• pyspark
• python
• sql