Hire me
Custom AI chatbot development with RAG
A chatbot for your website, help centre or internal knowledge base that answers from your own documents, says when it does not know, and hands over to your team.
Quick answer
I build custom AI chatbots with RAG (retrieval-augmented generation) for websites, help centres and internal docs: your content is indexed with embeddings and vector search, and the bot answers from it with the OpenAI or Anthropic (Claude) API, cites where the answer came from, and hands over to a person when it is unsure. For WhatsApp and Telegram, see my messaging bot service.
Free tool: calculate how much an AI chatbot would save you →
What I build
Website chat widget
A chat window on your site that answers visitors from your pages, docs and product data.
Internal knowledge assistant
A bot for your team over policies, wikis, tickets and documents, with access limited by role.
RAG pipeline
Document loading, chunking, embeddings and vector search, kept up to date when your content changes.
Answers with sources
Each answer can link to the page or document it came from, so users can check it.
Human handover and lead capture
Passes the chat to your team, or collects contact details and saves them to your CRM.
Quality checks and logs
Test questions before launch, conversation logs and rules for what the bot must never say.
Results on real products
Joni (joni.ai)
AI agents in production
I build the backend services, agent workflows and integrations of this AI product platform: the same building blocks a chatbot runs on.
WhatsApp AI support bot
Engineering article
I wrote up how I build a support bot on Claude and Node.js, with RAG, handover and safety rules.
Free RAG text chunker
Tool on this site
A free browser tool I built to split documents into chunks for RAG, the first step of every chatbot index.
Tools I use
RAGEmbeddingsVector searchOpenAI APIAnthropic APILangChainLlamaIndexNode.jsPythonReactMongoDBRedis
How a project works
- Scope call. We talk through the goal, the current state of the product and what success looks like.
- Written plan. You get fixed milestones, delivery dates and a clear price before any work starts.
- Build in small pieces. I ship in small steps with regular progress updates, so you always see what is built.
- Tested delivery. Delivery includes tests, documentation, handover and free support afterwards if something breaks.
Frequently asked questions
What is a RAG chatbot?
A chatbot that first retrieves the most relevant parts of your own documents and then has the model answer from them. It gives accurate, up-to-date answers without training a model.
How is this different from ChatGPT on our site?
It answers only from your content, follows your rules and tone, links to sources, logs conversations and hands over to your team. A plain ChatGPT widget does none of that.
What content can the chatbot learn from?
Website pages, help articles, PDFs, Word files, spreadsheets, Notion or Google Docs, and product data from your database or API.
Will the chatbot make things up?
It is built to answer from retrieved content and to say it does not know otherwise. Test questions before launch and logs after launch catch weak answers.
Can it work for internal company docs?
Yes. An internal assistant can answer staff questions from policies and wikis, with access limited by each user role.
Does it work in other languages?
Yes. It can answer in the user language, including English and Hebrew, even when the documents are in another language.
Do you also build WhatsApp chatbots?
Yes, on the official WhatsApp Business Platform and the Telegram Bot API. That is a separate service page on this site.
How much does a custom AI chatbot cost to run?
Mostly model API usage, which depends on traffic and answer length, plus hosting. Use the free AI chatbot ROI calculator to compare it with what support costs you now.
How long does it take to build?
A website chatbot over a clear set of documents often takes one to three weeks, including testing. Larger internal assistants take longer.
Who owns the chatbot and the data?
You do. The code, the index and the conversation logs stay in your accounts.