AI Agents
AI agents thatactually close.
Not a toy. Production-grade agents connected to your data and tools, with metrics and an admin panel.
Tech stack
- OpenAI / GPT
- Anthropic / Claude
- Google Gemini
- Function calling
- Vector DB (pgvector / Pinecone)
- Firestore
- AmoCRM API
- Telegram Bot API
- Wappi (WhatsApp)
- Node.js / Express
What I build
RAG assistants
A chat that answers from your knowledge base — without hallucinations. Citations, sources, fallback to a human.
Sales bots
Lead qualification with function calling, scenario classification, automatic pipeline movement in your CRM.
Process automation
Daily lead pulls, email classification, on-brand content generation, report aggregation.
Integrations
AmoCRM, HubSpot, Pipedrive, Telegram, WhatsApp, Instagram DM, Slack, Google Workspace.
Admin & metrics
Not a black box. Dashboard with dialogs, tags, response rating, exports and manual overrides.
Model-agnostic
GPT-5 / Claude / Gemini — we pick per task and budget. Swap without rewriting the project.
Questions about AI
Real. Function calling, tools, RAG, dialog classification, backend logic, integrations — what separates a working agent from a chat wrapper.
Depends on volume. A typical sales agent on GPT-4o-mini / Gemini Flash sits around €30–80/month in API at 1000+ dialogs. Full estimate on the discovery call.
Almost never in a well-built RAG. I enforce a strict-citation mode: if the answer isn't in the corpus, the agent says 'I don't know' and routes to a human. Controllable.
Through the native API. AmoCRM v4, HubSpot, Pipedrive — webhooks and REST. The bot tags, moves through the pipeline, leaves notes. Everything a manager does manually — automatic.
Plug it in. The architecture is provider-agnostic through an abstraction layer. Swapping GPT for Claude or Gemini is a day's work, not a week.