RAG-Powered PDF QA Platform
Budget / SalaryHourly project
TypeFreelance project
LocationRemote
Posted1 hour ago
I’m building a full-stack, AI-driven platform that lets people drop a PDF into the browser, log in with a simple username/password, and immediately start asking factual or concept-clarification questions about its content. Behind the scenes the service must:
• Ingest each PDF, split the text into manageable chunks, create sentence-transformer embeddings, and store them in a vector database (Chroma or Qdrant are both fine).
• On every question, perform a semantic search, pass the retrieved context into an LLM, and return an answer through a Retrieval-Augmented Generation workflow.
• Expose the whole pipeline through a clean FastAPI REST API with two core endpoints: /upload (POST) and /query (POST).
• Provide a lightweight Streamlit front end that handles sign-in, file upload, and chat-style Q&A, then streams answers back to the user.
• Include robust error handling, structured logging, and clear configuration so the service can be deployed straight to a standard cloud environment.
Technology stack must stay in Python and use LangChain (or equivalent custom orchestration), FastAPI, Sentence Transformers, Chroma/Qdrant, and any reputable LLM API.
Acceptance criteria
1. Upload a multi-page PDF and confirm embeddings are stored in the vector store.
2. Authenticated user asks a factual or concept question and receives an answer referencing the correct page range.
3. Response time for a typical 10-page document query is under 8 seconds on a modest GPU-free server.
4. Codebase is clean, fully commented, and includes a README covering setup, environment variables, and how to swap vector stores or LLM providers.
If this lines up with your expertise, I’m ready to review your approach and timeline.
• Ingest each PDF, split the text into manageable chunks, create sentence-transformer embeddings, and store them in a vector database (Chroma or Qdrant are both fine).
• On every question, perform a semantic search, pass the retrieved context into an LLM, and return an answer through a Retrieval-Augmented Generation workflow.
• Expose the whole pipeline through a clean FastAPI REST API with two core endpoints: /upload (POST) and /query (POST).
• Provide a lightweight Streamlit front end that handles sign-in, file upload, and chat-style Q&A, then streams answers back to the user.
• Include robust error handling, structured logging, and clear configuration so the service can be deployed straight to a standard cloud environment.
Technology stack must stay in Python and use LangChain (or equivalent custom orchestration), FastAPI, Sentence Transformers, Chroma/Qdrant, and any reputable LLM API.
Acceptance criteria
1. Upload a multi-page PDF and confirm embeddings are stored in the vector store.
2. Authenticated user asks a factual or concept question and receives an answer referencing the correct page range.
3. Response time for a typical 10-page document query is under 8 seconds on a modest GPU-free server.
4. Codebase is clean, fully commented, and includes a README covering setup, environment variables, and how to swap vector stores or LLM providers.
If this lines up with your expertise, I’m ready to review your approach and timeline.
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.