Browse categories
Explore
Fiverr Pro
English
$
USD
I will diagnose and fix rag retrieval and PDF extraction in python
Mangesh W
About this gig
Your RAG app finds documents, but misses the answer, loses PDF structure or gives weak citations?
I trace the failing path through extraction, chunking, retrieval, context assembly and source links. You get a reproducible diagnosis before committing to a larger rebuild.
My personal research assistant covers 499 papers and ~8M processed corpus tokens. Its developer traces and graph-guided retrieval are shown in the gallery; these are project examples, not promised gains for your app.
BASIC $100: review one Python pipeline, 2 sample PDFs/20 total pages and 5 failing questions. Written findings, prioritized fix plan and 30-minute review. No code changes.
STANDARD: Basic plus one agreed extraction or retrieval fix, source patch and 5 regression checks.
PREMIUM: up to 3 related fixes in that pipeline, 10 checks, source patches and a 60-minute handover.
One clarification revision for Basic/Standard; two for Premium. We agree the bug, setup and acceptance checks first. Full rewrites, new apps, large-corpus migration and complex OCR need custom scope. API/hosting fees are separate.
Message me with your stack, 2 sample PDFs and the failing questions.
Get to know Mangesh W
Mangesh W
RAG and Voice AI Developer
- FromIndia
- Member sinceSep 2024
Languages
English
I build AI applications that turn documents into answers with source citations, including relevant tables and diagrams. My work includes a research assistant over 499 papers, voice-enabled learning tools, and AI lecture-generation workflows. I work with Python, FastAPI, Next.js, React, PostgreSQL, LangGraph, Qdrant, and Neo4j. I can help with RAG chatbots, AI integrations, and full-stack development, with clear scope, working demos, and handover documentation.
My Portfolio
FAQ
Does the $100 package include a code fix?
No. Basic delivers a reproducible diagnosis and prioritized fix plan for one pipeline, 2 PDFs/20 pages and 5 questions, plus a 30-minute review. Standard adds one agreed fix; Premium adds up to three related fixes.
Can you guarantee perfect answers or a specific latency?
No. We agree on representative failing questions and check the results before and after an in-scope fix. Model, corpus and hosting choices affect behavior. The 8M-token showcase is evidence of my work, not your package capacity or an accuracy guarantee.
Which systems can you review?
Python RAG systems using FastAPI, LangChain, LangGraph, LlamaIndex and vector databases such as Qdrant. Send the relevant stack and setup notes first so I can confirm fit. This gig is for an existing pipeline, not a new application.
What access and costs are required?
Provide a minimal reproducible example, sample documents and sanitized logs. Do not paste secret keys or passwords in chat. We agree an appropriate access method if needed. API, model, database and hosting usage fees are separate.
