YouTube Video
Project Description
Akechi is a voice-first research assistant where users speak to explore topics and watch research cards appear on an infinite canvas. The system uses pgvector semantic search to automatically link related concepts and features a neuron-inspired memory system with activation decay and spreading activation, accessing one concept boosts connected ideas, mimicking human recall. The research part is handled through perplexity sonar which returns the links to sources it finds that the user can visit manually to verify information or read more in depth.
Tech stack: Next.js 16, ElevenLabs Conversational AI, Supabase (pgvector + Realtime), OpenAI embeddings, Perplexity Sonar Pro for research, Google Gemini Flash for diagram generation, React Flow for the canvas.