Browse categories
Explore
Fiverr Pro
English
$
USD
I will build a rag pipeline over your documents
Michael S.
About this gig
A RAG system that returns confident wrong answers is worse than no RAG at all. The work sits in chunking, retrieval quality and citations, not in calling an embedding API.
What you get:
- Document ingestion with chunking that respects how your documents are actually structured
- A vector store set up and indexed, with metadata filtering
- Retrieval with reranking, so the right passage reaches the model
- Answers with citations back to the source document
- A small evaluation set, so you can measure retrieval quality instead of guessing
I led an AI SaaS platform for 1.5 years with a team of 9 to 11 people using the OpenAI API, the Claude API and LangChain, with 18 years in backend engineering behind the infrastructure side.
I work from Ukraine on CET.
Send me a sample of your documents and the questions your users will ask. I will tell you which package fits and flag anything that will not work well before you order.
Get to know Michael S.
Michael S.
CTO and AI Developer, 18 Years: LangChain, RAG, Voice AI, Python Backends
- FromUkraine
- Member sinceFeb 2023
- Avg. response time1 hour
Languages
Russian, Ukrainian, English, German
I'm Michael, founder and CTO at Meduzzen, 18 years in backend and AI engineering. Most AI projects don't fail on the API call. They fail on latency, hallucination and cost once real users arrive. My team and I build the part that holds up: FastAPI backends, LangChain and RAG, Whisper and ElevenLabs voice, voice agent audits, n8n automation and full stack AI SaaS. Earlier, I worked on a 5G VoIP platform at Siemens, IoT security for WeWork, and blockchain threat detection at Cyvers. Send your repo or workflow and I will tell you honestly what it takes.
My Portfolio
FAQ
1. What document formats do you support?
PDF, Word, HTML, Notion exports and plain text. Send me a sample and I will confirm before you order.
2. Which vector database do you use?
Pinecone, Weaviate, pgvector or FAISS. I pick per project based on your scale, budget and whether you want it self hosted.
3. Can the RAG run on my own infrastructure?
Yes. I can deploy it on your servers with a local model if your data cannot leave your environment.
4. How do you make sure answers cite the right source?
Every answer carries citations back to the source document, and reranking puts the right passage in front of the model before it answers.
5. How many documents can it handle?
Up to 100 on Basic, up to 1,000 on Standard. Above that I set up incremental reindexing on Premium, or we scope it as a custom order.
6. Do you provide an evaluation of retrieval quality?
Yes. You get a small evaluation set so you can measure whether a change made retrieval better or worse, instead of guessing.
7. Can you connect it to my existing chatbot or app?
Yes. You get a query endpoint on Basic and a chat endpoint from Standard, so your app can call it directly.
