Browse categories
Explore
Fiverr Pro
English
$
USD
I will build a rag chatbot that cites your documents
Alexandr S
About this gig
Your answers already sit in your documents: return policies, manuals, price lists, help-center articles. People still ask the same questions, and a general chatbot invents answers. The chatbot I build searches your files first and answers from them, citing the file and page or section. When the documents don't cover a question, it says so.
Basic indexes up to 25 files (text PDF, Word, plain text or web pages) behind a simple web chat, with a guide to run it on your host. Standard takes up to 200 files, adds a chat widget for your site or a Telegram bot and chat history, installed on your server. Premium goes inside your Postgres or Supabase app with logins and access rules, or runs on your server with a local model, so files stay on your network.
Each package ends with a test on up to 20 of your questions: each answer, its source and whether it was right, then one round of fixes and a re-run. Source code is included. The bot runs on your own OpenAI, Claude or Gemini key, with no monthly fee to me.
Recent work: the vector search layer of a live Supabase backend for an election trading desk.
No calls. We agree the scope in the chat, and I report there each working day.
Get to know Alexandr S
Alexandr S
Full Stack Developer: Rust, TypeScript, JavaScript, AI Integrations
- FromKazakhstan
- Member sinceJun 2024
- Avg. response time1 hour
Languages
Russian, English
Full-stack developer with 10+ years in IT. I build and fix web, desktop and mobile apps, APIs and Chrome extensions in TypeScript, JavaScript and Rust, and add AI features: LLM integrations and RAG chatbots that answer from your documents. Stack: React, Next.js, Node.js (Express, NestJS), Rust (axum), PostgreSQL, MySQL, MariaDB, MongoDB, Redis, Supabase, WordPress, WooCommerce, Tauri, React Native, Docker. I also fix and secure apps built with Lovable, Bolt or Cursor that aren't ready for real users. I work async and in writing, with no calls.
My Portfolio
FAQ
Will it make answers up?
It answers only from the passages it finds and cites the file and page or section. If nothing relevant turns up, it says the documents don't cover the question. No model is perfect, so every package ends with a test on your own questions.
Which files can you use?
Text PDFs, Word files, plain text, Markdown and public web pages. Scanned PDFs need text recognition first, which is the OCR extra; handwriting is not covered. Spreadsheets work when each row reads like a sentence; send a sample and I'll tell you.
How do you count files?
One file is a document of up to 50 pages, or up to 25,000 words if it has no pages. A longer one counts as one file per 50 pages started. One web page is one file.
Where does the chatbot run?
On a server or host in your name, so you pay the host directly and owe me no monthly fee. Basic comes with a guide to run it yourself, or add the setup extra. On Standard and Premium, I install it on your server with the access you give me.
Who pays for the AI usage?
You pay the provider directly: the bot runs on your own API key, so every request shows up on your account. A public widget gets a per-visitor limit, so nobody can drain the key. The cost depends on how many questions people ask, so I don't promise a fixed number.
Is my data private?
The search index is built by a small model that runs with the bot, so your files don't go to the AI provider. Each question sends only the matching passages, under your account's terms. Premium can also answer with a local model on your server; that needs enough memory and is slower.
Can it go inside my existing app?
Yes, on Premium. If your app runs on PostgreSQL or Supabase, the search lives in your own database with pgvector, and access rules decide which user sees which documents. For other stacks, send me the details before ordering.
Do you do calls?
Calls aren't part of how I work. The order chat keeps a record of everything we agree.
