Mutemo Copilot

AI-powered legal aid that helps everyday Zimbabweans understand their rights, build cases, and generate print-ready legal letters — in English and Shona — powered by Amazon Nova.

Comment

Inspiration

In Zimbabwe, over 80% of the population cannot afford a lawyer. Workers go unpaid, tenants face illegal evictions, and domestic workers are exploited—but most people don't know their rights or how to take legal action. The few legal aid organisations that exist are overwhelmed, and the barrier isn't just cost—it's language, literacy, and knowing where to start. We asked: what if AI could be the legal aid office that's always open, speaks your language, and puts a real legal letter in your hands? That's Mutemo Copilot.

What it does

Mutemo Copilot guides users through the full journey from "I have a problem" to "here's your letter, ready to deliver":

  • Ask Mutemo: A conversational legal assistant grounded in the Constitution of Zimbabwe (English/Shona).
  • Case Builder: Users describe their situation in their own words; the AI extracts structured case data, timelines, and relevant legal provisions.
  • Letter Generator: Generates professional, print-ready legal letters (demand letters, complaints, grievances) with zero [PLACEHOLDER] fields.
  • Document Extraction: Users snap photos of contracts or notices, and the AI extracts data to pre-fill their case files.
  • Voice Assistant: Powered by Amazon Nova 2 Sonic for speech-to-speech interaction, ensuring accessibility for all literacy levels.

How we built it

  • Frontend: Ionic/Angular PWA deployed to S3 + CloudFront.
  • Backend: Node.js on AWS Lambda and Amazon Cognito for a stateless, serverless architecture.
  • Amazon Nova Models: We utilized Nova 2 Lite for case extraction, Nova Pro v1 for complex Shona translation, Nova Multimodal Embeddings for our RAG pipeline, and Nova 2 Sonic for real-time voice.
  • Hybrid RAG: We built a custom retrieval system over the Constitution using semantic search and re-ranking to ensure legal accuracy.

Challenges we ran into

  • Linguistic Nuance: Moving beyond "textbook" Shona to natural Zimbabwean code-switching required a custom two-step "Structure-then-Style" pipeline.
  • The "Placeholder" Battle: To ensure letters were truly ready to print, we built an agentic loop that proactively asks for missing info rather than leaving blanks.
  • Serverless Streaming: Managing bidirectional HTTP/2 streaming for Nova Sonic required a custom Socket.IO proxy to work alongside AWS Lambda.

Accomplishments that we're proud of

  • Constitutional Defense: Building a system that can "see" an informal eviction note and automatically flag it as a violation of Section 74 of the Constitution.
  • Zero-Friction UX: Successfully reducing the "Context Tax" by using multimodal extraction to fill out 90% of a legal form from a single photo.
  • Natural Shona Output: Achieving high-standard legal Shona that sounds authentic to local users.

What we learned

  • Low-resource languages require style-guided prompting and corpus references to avoid repetitive or nonsensical output.
  • Hybrid RAG is essential for legal text where exact terminology (lexical) matters as much as the general meaning (semantic).
  • Agentic Workflows solve the "Context Tax" problem better than traditional forms by inferring data from multimodal inputs.

What's next for Mutemo Copilot

  • Expanding Legal Domains: Adding support for the Labour Relations Act and Small Claims Court.
  • SMS Integration: Reaching users with feature phones through an SMS-based interface.
  • Pro-Bono Network: Creating a seamless hand-off from AI-generated files to human lawyers for courtroom representation.

Built With

Share this project:

Updates