Sakhi AI | Multi-Agent Safety Companion
Overview
Sakhi AI gives domestic violence survivors private, immediate access to legal information, safety planning, and support resources. Instead of one general chatbot, a LangGraph router sends each message to a specialist agent with its own focused instructions and tools. Because the stakes are high, every message is checked for risk first, so emergencies are never handled like an ordinary question.
Key Use Cases
- ✓Accessible information on domestic violence law and legal rights
- ✓Personal safety planning
- ✓Finding nearby NGOs, shelters, and helplines
Technical Insights
Designed a LangGraph multi-agent system where a router sends each message to specialist agents for legal questions, safety planning, and NGO, shelter, and helpline lookup via tool calling.
Built a 0–10 criticality scoring step that flags high-risk messages and surfaces emergency resources before normal routing.
Added cross-session memory using LangGraph checkpoints in PostgreSQL, so survivors don't have to retell their story, and RAG over domestic violence law (PWDVA) and support resources.
Currently building a voice pipeline (speech-to-text → LLM → text-to-speech) for hands-free use.