Service 04
OpenAI Realtime API Development
Speech-to-speech agents on the OpenAI Realtime API over WebRTC or WebSockets, with function calling, and an honest read on when speech-to-speech beats a chained pipeline.
What this service covers
The Realtime API removes two hops from the chain. Instead of transcribing audio to text, sending text to a model, and synthesising the reply back to audio, the model takes audio in and emits audio out. You get the lowest latency available today and a voice that keeps the tone and emphasis a transcript throws away.
You also give things up, and the trade is worth stating plainly. A chained speech-to-text, model and text-to-speech pipeline gives you a clean transcript at every turn, free choice of language model, cheaper token economics on long calls, and per-layer control you will want the first time compliance asks what the agent actually said. Speech-to-speech wins on latency and naturalness. Chained wins on control, cost and auditability. The right answer comes from the product, not from whichever is newest. I work through the trade in speech-to-speech vs chained pipelines.
I build both, and I have shipped both. That includes the Realtime specifics that catch people out: WebRTC versus WebSocket transport and when each is right, ephemeral token minting so an API key never reaches a browser, session and context management on long calls, function calling for real actions, and cost control on a billing model that charges for audio in both directions.
What you get
- A speech-to-speech agent on the Realtime API, over WebRTC or WebSockets
- Ephemeral token flow, so no API key is ever exposed to a client
- Function calling wired into your systems, with a human-approval path where it matters
- Session and context handling that keeps long calls coherent and affordable
- A written comparison of speech-to-speech against a chained pipeline for your use case
- Token and audio cost monitoring with alerting before the invoice, not after
How I work
Four stages, and you always know which one you are in
The same process whether the engagement runs two weeks or a year.
Understand the business
I ask product-level questions before quoting, so the work is scoped against what the business needs, not just the ticket as written.
Architecture in writing
You get the proposed architecture in writing before any code is committed. Disagreements happen on the document, where they are cheap.
Delivery in milestones
Work ships in milestones with regular updates, so you always know what has landed and what is next.
Documented handoff
Code is documented and handed off cleanly. Your future engineers should not have to reverse-engineer my decisions.
Direct Project Enquiry
Tell me what you are building
Share what you are building and where it is stuck. If a managed platform already solves it, I will say so. I review every enquiry personally and reply within about four hours.
- 1Send your project goal and the services you think you need.
- 2I review the context and whether I am the right fit.
- 3You get a direct reply with a practical next step.
International projects welcome
I work in English by email, video call or through Upwork, from GMT+7 with overlap hours for US and EU teams.
Read the Privacy Policy, Terms of Service and Cookie Policy.