NyayaSathi

Inspiration

We were inspired by the stark "justice gap" in rural India, where complex laws in English/Hindi alienate millions speaking local dialects. Witnessing how a lack of awareness leads to exploitation, we wanted to leverage AI not to replace lawyers, but to democratize basic legal awareness and create a bridge to the formal justice system.

What it does

NyayaSathi is a voice-first, multilingual legal aid agent. A user speaks their problem (e.g., "ওই জমি নিয়ে বিরোধ," "dahej ka jhagda") into their phone. Our AI classifies the issue, translates it, and speaks back a simple, actionable summary of their rights and next steps—all in their native dialect, bypassing literacy and language barriers.

How we built it

Frontend: Built with React Native for cross-platform mobile access.

Backend: FastAPI server to handle requests.

AI Core: Speech-to-Text: OpenAI's Whisper. Translation: Google Translate API & Facebook NLLB. NLP: Fine-tuned a Hugging Face Transformer model (e.g., BERT) on legal data for intent classification. Text-to-Speech: Google TTS.

Database: PostgreSQL to store user queries and legal resources.

Challenges we ran into

Low-Resource Languages: Finding quality training data for local dialects was a major hurdle. We combated this with data augmentation and leveraging pre-trained multilingual models.

Legal Accuracy: Ensuring the AI's guidance is factually correct and never harmful was paramount. We implemented a rigorous human-in-the-loop validation system for our knowledge base.

System Latency: Chaining multiple AI models (STT → Translation → NLP → TTS) introduced latency. We optimized with caching, model quantization, and asynchronous processing.

Accomplishments that we're proud of

We built a functional prototype that can accurately understand a spoken legal query in a local language and return a spoken, simplified response. Creating an intuitive, voice-based UI for a non-technical audience is our biggest achievement.

What we learned

We gained deep practical experience in the MLOps pipeline—from data collection for low-resource languages to deploying chained AI models. More importantly, we learned about the immense complexities and ethical responsibilities involved in building technology for sensitive domains like law.