Nyaya-Sathi

Inspiration

In India, the adage Justice delayed is justice denied is a lived reality for over 45 million pending cases. While participating in the Anthropic Hackathon, our team realized that the greatest barrier to justice isn't just the law, it's the Procedural Black Box.

Most litigants, especially the underprivileged and uneducated, have no idea why their cases drag on for decades. They see hearing dates come and go without knowing that specific rights, like Order XVII Rule 1 of the CPC (limiting adjournments to three), are being systemically ignored. We were inspired to build Nyaya-Sathi to move AI beyond a "chatbot" and into a "Procedural Auditor" that holds the system accountable.

How we built it

We adopted a modular, high-performance architecture designed to "compress decades of progress" into an 3-hour sprint.

The Architecture The Auditor (Deterministic Logic): We built a Python-based compliance engine that fetches real-time case data via judicial APIs. It audits case history against the Code of Civil Procedure (CPC) and Criminal Procedure Code (CrPC) using hard-coded legal timelines.

The Educator (Precedent-Aware RAG): We utilized Qdrant as our Vector Database, ingesting the OpenNyAI InJudgements dataset and 70 years of Indian Supreme Court judgments. We used InLegalBERT for specialized legal embeddings, ensuring the AI understands "Ratio Decidendi" over generic text.

Challenges we ran into

The 3-hour sprint was a battle against technical and ethical hurdles:

The Data hunt: It's quite difficult to find the real world and well patterned data for the official court judgements in India. We went through a straight 1 hr hunt for the data through various git hub projects and finally found one.

The Ethical Minefield: We wrestled with the risk of "Outcome Prediction." We made a conscious, mindful decision to refuse to predict if a user would "win" or "lose." Instead, we focused on Procedural Compliance, empowering users without providing biased or dangerous speculation.

Latency vs. Depth: Performing RAG over thousands of judgments while running OCR and Whisper in a single Streamlit app was a resource nightmare. We optimized this by using Groq for high-speed inference to ensure the user experience remained seamless.

Accomplishments that we're proud of

Real-World Utility: We didn't just build a chatbot; we built a system that generates a Legal Health Score. Seeing a CNR number turn into a list of actionable violations felt like we were finally turning the lights on in a dark room.

Deep Integration: We are incredibly proud of integrating the InLegalBERT model. It allowed our "Educator" to understand the difference between "Stay" and "Adjourn" in a way that generic models simply cannot.

Multilingual Access: We successfully implemented a translation layer that allows a farmer in rural India to ask a question in his native tongue and receive an explanation of the Supreme Court's precedents.

What we learned

We learned that the law is not just about "sections"; it’s about procedure. We spent hours diving into the dense text of the CPC and emerged with a profound respect for the complexity of the judiciary. Technically, we leveled up our skills in Vector Database management and modular system design, learning how to make disparate AI models (Whisper, BERT, Llama) talk to each other in a unified pipeline.

What's next for Nyaya-Sathi

Nyaya-Sathi is just beginning. Our roadmap includes:

Bhashini Integration: Moving beyond standard translation to the government-backed Bhashini API for superior vernacular accuracy.

Systemic Heatmaps: Aggregating anonymized audit data to show which districts have the highest procedural delays, creating public pressure for judicial reform.

Direct E-Filing: Partnering with legal-aid clinics to allow users to file their "Speed-up Petitions" directly through our platform.

We believe that by providing transparency, we can transform the legal system from a source of frustration into a source of flourishing. Justice shouldn't be a privilege of those who understand the jargon; it should be a right for everyone.