Browse categories
Explore
Fiverr Pro
English
$
USD
I will build a telegram ai chatbot that answers from your documents, with citations
Ivan Popovych
About this gig
Your people ask the same questions about the same documents - a policy, a contract set, a handbook - and the answer is always on page 40 of something.
This gig puts them behind a Telegram bot that answers in plain words and names its source: document, section, printed page. When the documents do not cover a question, it says so.
That refusal is the point: a confident wrong answer about your own policy is worse than no answer.
Documents are cut on their section boundaries, so a citation names a section rather than an area of a page, and every citation is checked against what was retrieved - an invented reference is dropped, not shown.
BASIC: one document set, answers with section citations.
STANDARD: printed page numbers, and a score on your questions.
PREMIUM: deployed on your hosting, with spending ceilings and analytics.
I built this on seven NIST publications, 1,019 pages, and scored it: 18/18 answers with no invented reference, 5/5 unanswerable questions declined, 11/13 answerable ones correct. Those are that build's numbers, not a promise about yours.
Tell me what your documents are and what your people ask, and I will say plainly if this is the right tool.
Get to know Ivan Popovych
Ivan Popovych
Software Developer
- FromUkraine
- Member sinceMay 2023
Languages
English, Ukrainian, Russian
I build automation and web apps that remove manual work: Telegram bots, Mini Apps, scrapers, API integrations, LLM chat over your documents.
For a property management firm I built an internal recruitment system on Telegram: a bot plus a Mini App with role-based access, notifications, interview scheduling, analytics. It had all been on paper.
Stack: Python, JavaScript, TypeScript, React, Node.js, SQL, REST APIs, Google Apps Script.
You get working code, a setup guide and written updates as I go.
Send me your task and I will tell you honestly if I can do it, how long and what it costs.
My Portfolio
Other Chatbot Development Services I Offer
FAQ
Will it invent answers?
That is the failure this is built against. Every citation is checked against the passages actually retrieved, and an invented reference is dropped rather than shown. On the build I scored: 18/18 answers with no invented reference.
What does it do when the answer is not in my documents?
It says so plainly. Refusing is a designed behaviour rather than an accident, and it is scored like everything else: on my build, all five unanswerable questions were declined, none of them answered anyway.
Where do my documents go? Do you keep a copy?
They become an index you own and keep. I hold no copy after handover. Embeddings run locally on the CPU, so indexing sends nothing anywhere; only a question and the extracts it retrieved ever reach the language model.
What does it cost to run once it is live?
Indexing costs nothing - it runs on your own machine. Answering costs model tokens, and the build carries two hard ceilings counted in the database, per person per day and per month, so a restart hands nobody a fresh allowance.
Can I change the documents later, or add more?
Yes. The document set is configuration, not code: point it at new files and reindex. On the build I scored, replacing all seven publications with another set is an edit to one config file.
I want a bot that takes orders or bookings, not one that answers questions. Can you do that?
Yes, but not with this gig - that is my other one, the Telegram bot and Mini App that replaces a manual process. It is rule-based: menus, forms and steps you define. Message me if you are not sure which of the two fits.
