Mikhail Semenov
RU

← Back home

Obrabot — internal voice AI analytics cabinet at Prof-IT

I run it as the PM: a personal cabinet with analytics and dashboards for the group's voice AI agents.

Role Product Manager
Timeline 2026, active group product
Key Impact Visibility into voice AI agent conversations for the team and customers

Obrabot: internal voice AI analytics cabinet

Obrabot is an internal product at the Prof-IT group. It’s a personal cabinet where customers and operators see what voice AI agents actually do: what they talk about, where the script breaks, and which calls need a human in the loop.

I run the product: shape the requirements, prioritize the backlog, align UX and data model with the team and engineers.

The class of problem

A voice AI agent without the right dashboard is a black box. The team can’t see:

  • which segments of calls go off-script;
  • where the AI agent performs worse than a human and why;
  • which specific calls a person actually needs to listen to;
  • how the conversation metric shifts from iteration to iteration.

Without this layer the product sells worse, because the customer doesn’t trust what they can’t verify.

What’s inside

  • AI call segmentation. Calls are automatically tagged by outcome, refusals, tone, and unusual behavior.
  • Dashboards by agent and script. Prompt version comparisons, scenario A/B, daily and per-campaign trends.
  • Drill-down to a specific call. Transcript, tags, audio, and metadata, all on one screen.
  • Early signals. Surfacing dialogs where a human should step in before the customer complains.
  • Prompt generation from client material. A registry of product prompts and a library of vetted references: the system assembles a new voice prompt out of the brief, the spec, and a call transcript.
  • Commercial proposal assembly. A short form describing the task, the niche and the integration, from which a proposal is generated with development cost and a per-second rate.
Obrabot prompts section: counters for generated and reference prompts, a form generating a voice prompt from PDF material, and the reference library per project (demo data)
Prompt registry: generation from client material on top of a reference library · demo data
Obrabot commercial proposals section: counters for ready and draft proposals and the generation form with task, model and integration pickers (demo data)
Commercial proposals: the form, the model picker, and rate maths · demo data

Stack

Next.js, TypeScript, Prisma + PostgreSQL, OpenAI API, integrations with Prof-IT infrastructure.

I’m not an “external AI consultant” talking about voice AI’s potential. I own an actual product inside the Prof-IT group, where voice AI is the main product line, not a pilot. That means real operational pain, real customers, NDA on the numbers, and accountability for outcomes.

The screenshots above come from the live cabinet on demo data. Real customers, their prompts and the numbers stay under NDA; happy to discuss details in a private conversation.