Sparr is a voice-first AI companion for conversations that matter. Users speak with an agent that listens, reasons, and replies in real time.
← All work· 2025
Sparr
Real-time voice AI for meaningful conversation. Built on LiveKit Agents, LangGraph, and a custom RAG pipeline.
Overview
The problem
Voice AI products tend to feel like dictation tools wired to a chatbot. The goal was something that actually held a conversation — low-latency, with memory, and useful context behind every response.
What we built
Live audio transport on LiveKit. Agent orchestration with LangGraph. A retrieval pipeline over a domain-specific corpus to keep responses grounded. Custom prompt and tool scaffolding to shape the agent's behavior across long sessions.
Outcome
A production voice agent with sub-second response latency and persistent conversational memory across sessions.
Stack
LiveKit Agents · LangGraph · Python · RAG · OpenAI · Postgres