Developer for AI platform - backend and API integrations
Hello everyone,
We need a developer for an AI platform — backend and API integrations
The B2C SaaS platform implies sessions lasting 25–35 minutes per user. The expected load at the start is 50–200 concurrent sessions. The architecture must withstand this without degrading quality.
---
What needs to be done
Six isolated technical tasks. Each is delivered separately with documentation.
---
**Task 1 — Streaming pipeline (voice + video in real time)**
Connect three external APIs into a single low-latency stream. Goal: the user finishes speaking — the system responds with voice and an animated face within 2–3 seconds. Everything works in parallel, not sequentially. Error handling and fallback are needed if one of the services does not respond. (Claude, ElevenLabs, Tavus)
---
**Task 2 — Session memory system and variations**
After each session, the system automatically extracts a structured summary: what was covered, where the user responded weakly. The summary is saved in the database. In the next session — it is intelligently compressed and embedded in context so that each new session differs from the previous ones: different topics, different angles, targeted work on weak points. The architecture should not break under 10–15 sessions per user.
---
**Task 3 — Automatic collection and processing of thematic information**
The user enters text or a title — the system automatically collects relevant public information from open sources and compresses it into a short contextual block. Edge case handling is needed: source unavailable, insufficient data, request not recognized.
---
**Task 4 — AI analysis of user content**
The user pastes arbitrary text — the system analyzes it through AI, returns a structured report with ratings by categories and specific improvement recommendations. The result is displayed in a ready-made UI.
---
**Task 5 — AI generator of personalized texts**
The user enters context and parameters — the system generates personalized text through AI. The ability to regenerate, edit, and download the result is needed. High usage frequency — the architecture should be cost-effective to operate.
---
**Task 6 — Dialogue simulator with increasing pressure**
A separate dialogue module based on the same voice/text stack as in Task 1, but with different scenario logic. The user enters initial conditions — the system conducts a simulation with gradually increasing pressure from the AI interlocutor.
---
Who we are looking for
- Confident in JavaScript / Node.js (or Python)
- Real experience with streaming and realtime APIs — not theoretical
- Understanding of asynchronous programming in practice
- Experience with external AI / voice / video APIs
- Clean code with documentation — you deliver not just "working" but understandable
---
Work format
- Fixed budget for results, not hourly
- Possible long-term cooperation as the product grows
---
When responding, please indicate:
1. A specific example of work with streaming or realtime APIs
2. Examples of similar integrations if available
3. Your estimate of timelines and costs for 2 stages: tasks 1-3, tasks 4-6.
Have a great day everyone!
-
21 days20 860 USD
1168 7 0 21 days20 860 USDHello! We have experience in creating scalable backend solutions for AI platforms with loads of 200+ sessions. We implement this through asynchronous architecture on FastAPI, WebSocket, and message queues. I guarantee stability, low latency, and easy integration with AI services.