Aigiz Kunafin
Production real-time Voice AI Agents.
Consulting: from ASR/TTS to agents that take actions and use RAG over your data — with control of quality, latency and cost.
UTC+5 · 30+ min · EN / RU
Homai at AI for Good
RAG evals / robustness
traffic spikes
I’ve been building production systems since 2011. I’ve been building applied ML systems since 2018 (LLMs and agentic systems in recent years).
My specialization is voice agents that run in real time and behave predictably in production: quality, latency, cost, observability, and robustness.
How I help products
Production voice
I help build voice pipelines where latency, quality and cost are controllable — and releases are predictable.
Agents that act
We design the agent layer: tools, workflows, routines and background jobs. Agents don’t just answer — they act, e.g., prepare a research brief by the start of the workday or analyze server logs and send a report.
RAG you can trust
We set up retrieval and evals so answers stay grounded in your data and robust to errors.
My specialization — I help teams with
- Voice agent MVP: pick the stack (ASR/TTS/LLM), architecture and pipeline, ship the first version fast.
- Real-time & quality: reduce latency, streaming setup, improve ASR/TTS, define metrics.
- Agent actions: tools and routines (automation, “do X every day”).
- RAG over your data: retrieval over docs/KB, evals, error control.
- Production: turn an MVP into a reliable solution: cost, reliability, observability, scaling under load.
- Production support: diagnose what went wrong, handle incidents, find root causes and stabilize the system.
Consulting
Call + practical next steps
We can do a quick call, or you can share context beforehand for a deeper review.
I can propose architecture options and help you choose the optimal path for your constraints, help pick models/tech, highlight pitfalls, review your pipeline and outline practical next steps.
To start faster, send a short message using this template:
Product: … Stage: … (idea / MVP / production) What you need: … (quality / latency / RAG / agent actions / architecture) Constraints: … (languages, platform, timeline, budget range if ok) Link/repo/demo (if any): …
Large companies: I’m not looking for full-time roles, but 1:1 and group consulting is possible.
Featured work
Homai
Homai (I’m the founder) — a reference implementation of a real-time voice agent: on-device wake word, ASR/TTS workers, an LLM agent with actions, and RAG over user data. (We also manufacture in-house: our SMD assembly machine runs software we wrote, using ML models and a vision LLM for quality control.)
If you’re building a voice product, this is a good reference for architecture and production pitfalls.
Other projects
AI for Good
Geneva, 2025 · Homai
Enterprise RAG Challenge 3
Accuracy leaderboard (#9)
YouTube translation & dubbing
ASR / diarization / translation / TTS
Language digitization
data → corpora → models
Want to ship a production voice agent?
Send 3–5 lines of context and I’ll suggest the next step: a quick call or a deeper review with prep.