Browse categories
Explore
Fiverr Pro
English
$
USD
I will build a rag chatbot over your company documents with source citations
Riccardo S.
About this gig
A chatbot that answers from your manuals, policies and procedures, and shows the passage it answered from so your team can check it.
Plain vector search fails on a large share of real questions: product codes, exact settings, follow-ups like "and for the bigger model?". It fails confidently, with a citation. I build assistants that look at the question first (what kind it is, what must match letter for letter, which document it is about) and I measure retrieval on your own questions before you accept the delivery.
WHAT YOU GET
- Ingestion of your PDF, Word and Excel files
- Chat with citations to the exact passage
- Hybrid search (meaning + exact keywords) on PostgreSQL with pgvector
- OpenAI, Anthropic, Mistral or fully local models
- An evaluation on your questions, with the results
- Full source code, Docker, README and a handover video
WHY ME
Full-stack engineer since 2017. I have built document assistants in production, and my portfolio measures exactly where plain vector search goes wrong.
Please message me before ordering so we can confirm scope and timing.
Get to know Riccardo S.
Riccardo S.
AI Integration Engineer for LLM Data Extraction, RAG and OCR
- FromItaly
- Member sinceSep 2026
Languages
Italian, English
I connect AI to the systems a business already runs: mailboxes, PDFs, ERPs, SQL Server and PostgreSQL databases. Full-stack engineer since 2017 (Node.js, Python, Angular, React, .NET). I build LLM extraction pipelines where every value points back to the words it was read from, document assistants measured on your own questions, secure OCR APIs and internal dashboards. Fixed scope in writing before any code; tests, Docker and a handover video with every delivery. CET time zone. Please message me before ordering so we can confirm scope and timing.
My Portfolio
FAQ
Do you work on healthcare or medical projects?
Yes. I have built medical imaging software (DICOM viewers, PACS integration) and patient portals. If your data includes patient information, we work on anonymized samples and agree on data handling in writing before any code.
Will it answer questions that aren't in the documents?
No, and that's deliberate. It answers from your documents, shows the passage, and says so when the documents don't contain the answer.
Can my documents stay private?
Yes. It can run entirely on your own server with local models, so no document leaves your network. If you prefer cloud models, I tell you exactly what is sent to the provider, and when, before we start.
Do you provide the API key?
No. You use your own OpenAI, Anthropic or Mistral key and pay its usage, or run local models at no per-question cost. I estimate the monthly cost before starting.
Can you deliver in 24–48 hours?
No. I work part-time, evenings and weekends CET, and the delivery times in the packages are real. If you need it faster, please don't order.
Do you use AI coding tools?
Yes, to work faster. Every line is reviewed and tested by me, and you get the full source. If your policy forbids AI-assisted code, tell me before ordering and I'll decline.
Who owns the code?
You do, once the order is complete. It is delivered as a Git repository.
