An AI assistant trained on your documents
Answers customers and staff 24/7 — in Telegram, on your website and inside your CRM. Knows your policies, price lists and manuals. Doesn't make things up: it answers from the knowledge base and hands complex questions to a human.
- 2–5 weeksfrom contract to launch
- 100,000–250,000 RUBfixed price by contract
- 24/7answers, no days off
Sound like your day?
An AI assistant removes the routine that eats managers' time and loses customers.
The same questions, again
Managers spend half the day answering "how much", "how do I pay", "how long" — instead of selling and solving hard problems.
Customers write in the evening
A request comes in at 10:40 pm, the reply — in the morning. By then the customer has bought from whoever answered in a minute.
Newcomers take forever to ramp up
A new employee pesters colleagues about policies and price lists for a month, because the knowledge is scattered across chats and folders.
A knowledge base nobody reads
The manuals are written, but nobody reads them: asking is faster. Documents go stale, mistakes repeat.
One assistant — two audiences
The external loop works with customers, the internal one — with your team. You can roll out both or start with one.
Answers customers
Prices, terms, timelines, availability — from your current price list and policies, not from the model's "memory".
Advises staff
Policies, manuals, document templates — a newcomer gets an answer in seconds instead of pestering colleagues.
Cites the source
Every answer links to the document and clause it came from. Easy to verify and trust.
Calls a human in time
Unsure of the answer or the customer asks for a manager — the assistant hands the conversation over with full context.
Collects leads into the CRM
Contacts and the gist of the request go to amoCRM, Bitrix24 or 1C — the manager sees ready leads in the morning.
Lives in your channels
A Telegram bot, a website widget, the company's internal chat — one brain, as many entry points as you need.
How it works inside
RAG architecture in plain words — no magic, no marketing.
We collect the knowledge base
Documents, price lists, policies, website pages, chat history. I help select and structure them — it's part of the project.
Search finds what's relevant
For every question, vector search pulls exactly those document fragments from the base that contain the answer.
The model phrases the answer
GigaChat, YandexGPT, ChatGPT or an open-source model on your server — answers from what was found and cites the source.
RAG (Retrieval-Augmented Generation) is an architecture where the model answers not from its "memory" but from your documents. That's why the assistant doesn't invent facts: if the base has no answer, it says so and passes the question to a human.
The assistant answers. The agent acts
The next level is an AI agent: it doesn't just answer questions, it performs routine operations in your systems. I implement it as an extension of the assistant or as a separate project.
Lead qualification
The agent clarifies the task, budget and timeline, creates the deal in the CRM and assigns a manager — before you open the chat.
Documents from templates
Proposals, contracts, invoices — the agent assembles them from your templates and deal data.
Inbox triage
Requests from email, forms and messengers are sorted, cards filled in and tasks assigned to the right people.
Digests and reports
A morning digest on sales, stock or requests — to the manager's Telegram, with no manual export.
Works inside your systems
amoCRM, Bitrix24, 1C, Google Sheets, calendars and payment services — the agent acts through their APIs.
Under control
The agent's permissions are limited; important actions require human approval. Every operation is logged.
An employee who has read all your documents, answers in seconds and never quits.
What it costs
A fixed price by contract — after the free AI audit and a brief. No hidden extras.
The price depends on the size and format of the knowledge base, the number of channels (Telegram, website, CRM), integrations and data-hosting requirements (cloud or your own server). Ongoing support — from 15,000 RUB/month, the first 3 months are covered by the warranty for free.
- Process audit and scenario design
- Knowledge-base collection and preparation
- RAG pipeline and prompt setup
- Integration: Telegram, website, CRM
- Testing on real customer questions
- Training for your team
- 3 months of warranty support
- Source code and access are yours
How the project goes
You try a prototype on your own documents within the first 1–2 weeks.
Free AI audit
We review the processes and find where the assistant pays off. If it won't — I'll say so.
Spec and contract
We fix scenarios, channels, quality metrics and the price. Sole proprietor, contract, NDA, staged payment.
Prototype and tuning
A working assistant on your documents. We test it on real questions and tune the answers.
Launch and training
We connect the channels, I train your team to work with the knowledge base. Then — warranty and support.
Related case studies
Projects with ML and customer-communication automation — with numbers and architecture. Case studies are in Russian.
AI pricing for a hotel chain
A dynamic-pricing ML model for 3 hotels (280 rooms). Result within 4 months.
Read the case (in Russian)A bot for a Wildberries seller
Order automation for a seller with 15M RUB/month turnover.
Read the case (in Russian)AI in business: 5 ways to automate
How to implement AI in customer service, sales, analytics and operations — with real metrics.
ReadFrequently asked questions
What business owners ask before implementing an AI assistant.
Won't the AI assistant make up answers?
No. The assistant is built on RAG architecture: it answers only from your knowledge base and shows the source of each answer. If the base has no answer, it says so honestly and hands the conversation to a human instead of inventing one.
Which documents work for the knowledge base?
Almost any: PDF, Word, Excel, Google Docs, Notion, website pages, policies, price lists, manuals, customer chat history. During implementation I help you collect and structure the base — it's included in the price.
How do we update the knowledge base after launch?
New documents are uploaded in a couple of clicks — the assistant starts using them immediately, no retraining needed. I set up automatic sync with your sources (for example, a document folder or the website) and train your team to work with the base.
What about data security?
I work under a contract with an NDA. For sensitive data I use Russian models (GigaChat, YandexGPT) with data stored in Russia, and when needed I deploy an open-source model on your own server — the data never leaves the company.
How is this different from a "bot with ChatGPT plugged in"?
A bot with ChatGPT answers from the model's general knowledge and easily makes things up. A RAG assistant answers from your documents with a link to the source, integrates with your CRM and hands complex questions to a human. It is a controlled system, not a chat with a neural network.
How much does it cost and what affects the price?
Turnkey implementation costs 100,000–250,000 RUB and takes 2–5 weeks. The price depends on the size and format of the knowledge base, the number of channels (Telegram, website, CRM), integrations and data-hosting requirements. I give an exact quote after the free AI audit.
What if the assistant answers poorly?
Before launch I test the assistant on real questions from your customers and tune the answer quality together with you. After delivery — 3 months of warranty support: fixes and adjustments are free.
How long does implementation take?
2–5 weeks from contract to launch. You see a working prototype on your documents within the first 1–2 weeks — and try it hands-on before the full rollout.
Find out whether an assistant pays off in your business
I review your processes and show where AI saves money — with an estimated effect and budget. If it won't pay off, I'll say so.