Powered by OpenAI

Intelligence that understands context.

Praxis isn't just a transcriber. It's an active participant that learns from previous sessions, understands medical/legal terminology, and drafts documents like a human expert.

Semantic Memory

Our Vector DB (Supabase pgvector) stores every session. Ask "What did we discuss about the patient's medication last month?" and get an instant answer.

Contextual Follow-ups

The AI suggests follow-up questions during the live session based on missing information required for the final report.

Zero-Shot Templates

Create new document types just by describing them. "Create a referral letter for checking cardiologist with a focus on arrhythmia."

Under the hood.

  • RAG Architecture

    Hybrid search combining keyword and semantic embeddings for perfect recall.

  • Real-time Streaming

    WebSockets push audio chunks to Deepgram and stream text back in <300ms.

Explore the Tech
vector_search.ts

const embeddings = await openai.createEmbedding(query);

const results = await supabase.rpc('match_documents', {

query_embedding: embeddings,

match_threshold: 0.78,

match_count: 5

});