AWS Builder Center
Why ChatGPT fails miserably at Indian village law (and how we built a 5-Agent RAG mesh on Amazon Bedrock that doesn't)

Why ChatGPT fails miserably at Indian village law (and how we built a 5-Agent RAG mesh on Amazon Bedrock that doesn't)

If you ask standard ChatGPT: "My neighbor encroached 2 feet on my wheat farm boundary in Varanasi, what should I do?"

CS Undergrad and AI/ML Developer
If you ask standard ChatGPT: "My neighbor encroached 2 feet on my wheat farm boundary in Varanasi, what should I do?"
before you read full article you can just go and checkout our github repo :  https://github.com/shankywho/ClearCase.git 
It will advise you to hire a high court lawyer, cite irrelevant central property laws, and suggest expensive litigation. That is not just unhelpful—it is financially catastrophic for a rural farmer.
Indian rural law is intensely local. Agricultural boundary disputes in UP are governed strictly by Chapter IV of the UP Revenue Code, 2006, while agricultural wage claims belong to Section 20 of the Minimum Wages Act, 1948.
In ClearCase, we engineered an agentic RAG mesh powered by Amazon Bedrock:
  1. Dialect Transcription Agent: Ingests raw audio in Bhojpuri, Awadhi, or Haryanvi, normalizing colloquial idioms without losing factual intent.
  1. Statutory Vector Matching Agent: Powered by Amazon Titan Embeddings (amazon.titan-embed-text-v1), it searches state revenue codes to find the exact sub-section governing the dispute in milliseconds.
  1. Panchayat Precedent Memory: Law in books must respect village tradition (reeti-riwaaj). This agent recalls past successful Gram Sabha resolutions so settlements feel culturally natural.
  1. Coercion & Exploitation Safety Gate: Before drafting anything, this gate intercepts predatory loans (greater than 36% compounding interest), seized identity cards, or caste boycotts, escalating them straight to District Legal Services Authorities.
  1. Multi-Model Resilience: Claude 3.5 Sonnet handles primary drafting, with automatic failover to GPT-OSS 120B on Groq, Qwen 2.5 32B, or our offline rule engine during network drops.
Engineered by Team TableTurners:
  • Akshat Arya (Team Lead): Designed the 5-agent RAG pipeline, prompt guardrails, and statutory semantic chunking from India Code.
  • Ansh Johnson: Integrated real-time audio streaming and visual confidence indicators for the AI outputs on the frontend.
  • Shankar Pratap Singh: Built the resilient microservice bridge between Node.js and Python FastAPI with sub-second health checks.
  • Mayank: Structured the deterministic JSON outputs into cryptographically signable payloads for on-chain anchoring.
59 automated tests passing across our test suite. In legal tech, reliability is not a feature—it is a requirement.
#MachineLearning #AmazonBedrock #MultiAgentSystems #RAG #ClaudeSonnet #Groq #Python #TableTurners
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article