Open / Wildcard

Vistara AI

Vistara AI: See What Matters

Demo video

What it does

Vistara AI helps people understand contracts before they sign them. A user uploads a PDF agreement or pastes contract text. Vistara extracts the content, analyzes it using AI, and returns an easy-to-read contract safety score from 0 to 100. It identifies potentially risky clauses, labels them High, Medium, or Low risk, explains them in plain English, and describes possible worst-case consequences. Most importantly, Vistara connects risk findings back to the original document, helping users see where the concerning clause appears instead of relying on a vague AI summary. Vistara is not a replacement for legal advice. It is a practical first step for renters, landlords, freelancers, job seekers, and anyone who needs to understand an agreement before making an important decision.

Inspiration

Vistara began with Jahanvi’s experience with rental agreements. When Jahanvi rented a place, she was expected to sign a long agreement filled with legal language she could read but could not confidently interpret. She did not know whether it contained hidden responsibilities, unfair conditions, financial penalties, or terms that might create problems later. Later, when she rented a place to someone else, she saw the issue from the other side. She wanted an agreement that protected her, while also making sure it did not unintentionally harm the tenant. That raised a question for our team: why should understanding an important contract require legal expertise or the money to hire a lawyer? We built Vistara to give regular people a clearer starting point before signing.

How we built it

We built Vistara AI as a cross-platform Flutter application with a FastAPI backend. The platform accepts PDF and text contracts. The backend validates and extracts document text with pdfplumber, cleans and segments it into clauses, and sends structured content to Google Gemini for analysis. Gemini identifies risks, assigns severity levels, generates explanations in plain language, and describes potential worst-case outcomes. The Flutter app presents the results through a clean risk dashboard, an overall safety score, detailed clause explanations, document viewing, and local storage for recent analyses. Team contributions: - Ummulkiram — Vistara Extract: PDF ingestion, validation, text extraction, preprocessing, clause segmentation, and structured contract data. - Jahanvi — Vistara Intelligence: Google Gemini integration, prompt engineering, risk detection, severity scoring, plain-language explanations, worst-case scenarios, and recommendations. - Hiral — Vistara Insight: Risk visualization, clause-to-risk mapping, PDF highlighting, dashboards, and the decision-support user experience.

Challenges we ran into

Our hardest challenge was making Vistara point to the risky language inside the PDF itself. At first, AI could identify a concern and return an explanation. But that was not enough. If a user sees “High Risk,” they should not have to search through several pages trying to find the relevant sentence. They should be able to see the clause, read it in context, and understand why Vistara flagged it. Making that happen required fuzzy matching between extracted contract text and the original PDF. This was challenging because PDFs do not always preserve text consistently: formatting changes, text can be split across lines or pages, and the same words may appear in more than one place. None of us had built this kind of document-highlighting workflow before. We had to learn text extraction, clause mapping, and fuzzy matching while still building the rest of the application during the hackathon. We also faced API bugs that interrupted the analysis flow, and resolving them was a key moment that allowed Vistara to become a complete working product. What we learned Vistara taught us how to build a complete cross-platform AI product—not just an AI demo. We learned how to connect Flutter and FastAPI, process real PDF documents, structure LLM prompts for reliable outputs, and design an interface that explains AI findings instead of overwhelming users with them. We also learned an important lesson about legal AI: detecting a problem is only the first step. People need to see the original clause, understand it in ordinary language, and know why it matters. That is why Vistara focuses on explainability and visual context rather than just generating a contract summary.

What's next

Next, we want to add: - Multilingual analysis - OCR for scanned or photographed agreements - Voice-to-text interaction - Offline availability and stronger privacy controls - A specialized contract-analysis model - A community forum for guidance and shared learning - A Vistara browser extension to review online agreements before users accept them Our goal is simple: no one should have to sign an important contract without understanding what it could mean for them.