This repo demonstrates an AI Agent prototype that mimics an MCP-backed system using FastAPI and Gradio, powered by llamaedge framework. It features contextual follow-ups, intellegent prompt routing and a chat-style frontend via a local JSON QA bank.
Uses Gradio to create a real-time conversation panel featuring:
a. Practice Question button that generates random questions, answers and explanation via the random_question() endpoint.
b. Ask Custom question button that receives a question entered by the user and fetches answer and explanation via the get_question_and_answer() endpoint.
c. Follow-Up for any other conversation based on the session context.
Custom Questions are answered using LLaMA2 via Ollama with responses managed per user session (UUID based).
Enables context-aware follow-ups using TF-IDF cosine similarity. Keyword-based prompts like "answer" and "explain" default to the latest interaction, while complex queries trigger similarity search over session history to identify the most relevant context for coherent multi-turn responses.
Contains 60 MCQs along with answers and explanation stored in a json file for the AWS Certification Programme prep.
A simple diagram explaining the system architecture:
The context switching endpoint implemented in the pretest using sentence-transformers works only when called after the get_question_and_answer() function.
Thank you for your time. I am excited to learn and grow through this mentorship opportunity. I'd be happy to improvise based on your feedback or suggestions!