BharatAid Copilot

Find the right government scheme, check eligibility, and get step-by-step help in your language.

Inspiration

We built BharatAid Copilot because finding the right government scheme in India can feel overwhelming, even when the support already exists. The information is scattered across PDFs, portals, and official notices, and it is often hard to understand in plain language or in your preferred language. We wanted to make access to public welfare feel less like paperwork and more like a helpful conversation.

What it does

BharatAid Copilot helps users discover relevant government schemes, check whether they might be eligible, and get a clear application checklist for the next steps. Users can ask questions in natural language, browse a searchable scheme catalog, and receive answers with citations from official sources. It also supports multilingual input and output so more people can use it comfortably.

How we built it

We built the frontend with Next.js 14, TypeScript, Tailwind CSS, and shadcn/ui, and the backend with FastAPI, SQLite, SQLAlchemy, and Pydantic. On the Databricks side, we ingested scheme documents into Delta Lake, used Spark for processing, Vector Search for retrieval, and MLflow to track experiments and evaluation. We also added SentenceTransformers, Transformers, and Playwright to support multilingual retrieval, demo validation, and a polished end-to-end experience.

Challenges

One of the hardest parts was dealing with messy, inconsistent government source documents and turning them into something searchable and trustworthy. We also had to balance multilingual support with retrieval quality, while keeping the answers grounded and easy to understand. Making the system feel fast and useful end-to-end, not just technically impressive, took a lot of iteration.

Accomplishments

We’re proud that BharatAid Copilot is not just a chatbot, but a full workflow for scheme discovery, eligibility checking, and application guidance. The source-grounded answer cards and citations make the experience feel more trustworthy, and the evaluation dashboard helps us measure quality instead of guessing. Most importantly, we turned a very real public access problem into a demo that feels practical and usable.

What we learned

We learned how much structure matters when building AI for real-world public information. Retrieval quality, source cleanliness, and clear UX are just as important as the model itself. We also got a lot better at using Databricks as a full platform for data ingestion, search, evaluation, and iteration.

What's next

Next, we want to expand the scheme corpus, improve language coverage, and make eligibility matching even more precise with better rule extraction. We’d also love to add voice input, more regional language support, and a deployment path that could actually help citizens outside the hackathon demo. Long term, we want BharatAid Copilot to become a reliable civic assistant for everyday people.

Built With

• databricks-delta-lake
• databricks-spark
• databricks-vector-search
• docker
• fastapi
• lucide-react
• mlflow
• next.js-14
• playwright
• pydantic
• python-3.11
• sentencetransformers
• shadcn/ui
• sqlalchemy
• sqlite
• tailwind-css
• transformers
• typescript
• zod