# Deepgram Flux Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/deepgram-flux
> Markdown URL: https://aitinkerers.org/technologies/deepgram-flux.md
> Technology record last updated: 2026-07-02T05:45:59Z
> Generated: 2026-09-22T07:36:21Z

Deepgram Flux is the first Conversational Speech Recognition (CSR) model: it embeds turn-taking intelligence directly into the model for ultra-low latency, human-like voice agents.

Flux is a streaming-first CSR model, engineered specifically for real-time voice agents. Unlike traditional ASR, Flux doesn't just transcribe: it understands the flow of dialogue, providing embedded turn-taking intelligence and native barge-in handling (interruptions). This eliminates complex client-side logic and delivers lightning-fast performance, with end-of-turn detection clocking in at ~260ms. You get Nova-3 level accuracy alongside this conversational intelligence, supporting enterprise scalability with 100+ concurrent streams per GPU. Flux provides the foundation for truly responsive, human-like voice AI.

- Official technology site: https://deepgram.com/flux
- Public AI Tinkerers demos and talks: 2
- Result page: 1 of 1

## Recent Public Talks and Demos

### [The Prompt That Worked in Isolation and Died in Production: Eval-Driven Prompt Engineering for a Real-Time Interview Copilot](https://seoul.aitinkerers.org/talks/rsvp_S4LRW_RGb3U)

InterviewMate (interviewmate.tech) is a real-time interview copilot I built and run in production: it listens to a live interview, transcribes it as it happens, and streams back answer suggestions grounded in the candidate's own experience — in under a second. The pipeline is Deepgram streaming STT into a FastAPI WebSocket Q&A pipeline, RAG retrieval over my own past answers, and a Claude streaming response. I'll run the working system live. Then I'll show the part that actually mattered: the eval harness behind the system prompt. It started as a bug — my copilot answered the viral "car wash" question correctly ("drive") while the same model, called with no prompt around it, said "walk," and I couldn't explain which prompt layer was responsible. So I turned the production prompt into a controlled experiment: 6 prompt conditions, 20 runs each, showing STAR-style structured reasoning beating context injection by 2.83x — and a follow-up where that same winning prompt collapsed to 0–30% once I dropped it into the real 60-line production prompt. Code, eval logs, and the pipeline architecture will all be on screen.

- Event context: AI Tinkerers Seoul: Inaugoral Demo Night — 2026-07-02 — Seoul
- Public talk page: https://seoul.aitinkerers.org/talks/rsvp_S4LRW_RGb3U

### [Building a Real-time Voice Agent with Cloudflare's Edge Stack](https://montreal.aitinkerers.org/talks/rsvp_FZMBRFtgd6Q)

Live demo of a sales pitch coaching app built entirely on Cloudflare: Durable Objects for stateful WebSocket sessions, Workers AI for real-time transcription and LLM inference, Vectorize for RAG, and R2 for document storage. I'll walk through the architecture decisions, show how structured extraction beats naive RAG chunking, and demo the app coaching a live pitch with realtime insights and post-session scoring.

- Event context: AI Tinkerers Montreal - January 2026 Meetup at Ateko — 2026-01-21 — Montreal
- Public talk page: https://montreal.aitinkerers.org/talks/rsvp_FZMBRFtgd6Q

## Related Technologies

- [Claude Sonnet 4](https://aitinkerers.org/technologies/claude-sonnet-4) ([Markdown](https://aitinkerers.org/technologies/claude-sonnet-4.md)) — 18 public demos
- [Cloudflare Workers](https://aitinkerers.org/technologies/cloudflare-workers) ([Markdown](https://aitinkerers.org/technologies/cloudflare-workers.md)) — 22 public demos
- [Durable Objects](https://aitinkerers.org/technologies/durable-objects) ([Markdown](https://aitinkerers.org/technologies/durable-objects.md)) — 8 public demos
- [FastAPI](https://aitinkerers.org/technologies/fastapi) ([Markdown](https://aitinkerers.org/technologies/fastapi.md)) — 181 public demos
- [Llama 3](https://aitinkerers.org/technologies/llama-3) ([Markdown](https://aitinkerers.org/technologies/llama-3.md)) — 38 public demos
- [OpenAI text-embedding-3-small](https://aitinkerers.org/technologies/openai-text-embedding-3-small) ([Markdown](https://aitinkerers.org/technologies/openai-text-embedding-3-small.md)) — 2 public demos
- [Qdrant](https://aitinkerers.org/technologies/qdrant) ([Markdown](https://aitinkerers.org/technologies/qdrant.md)) — 17 public demos
- [Vectorize](https://aitinkerers.org/technologies/vectorize) ([Markdown](https://aitinkerers.org/technologies/vectorize.md)) — 2 public demos