Enhanced Feedback Analysis System


Project Snapshots

System Architecture
1 / 8
System Architecture
Single Review - Input
NER Agent
Sentiment Quintuples
Coreference & Comparative
Batch - Upload
Batch - Insights
Batch - Summary

Project Technical Information

Project Name:
Enhanced Feedback Analysis System
Project Type:
Multi-Agent NLP Real-time SSE RAG
Tech Stack:
Python Flask (REST + SSE) Mistral + Phidata (Agentic) LangChain (RAG) FAISS all-MiniLM-L6-v2 Pandas Bootstrap Jinja2
Repository:
Repo: aiml_project_doc

Project Summary

A multi-agent NLP platform that extracts rich, structured insights from customer feedback. Beyond basic sentiment analysis, it performs NER, Sentiment Quintuple Extraction, Coreference Resolution, and Comparative Opinion Mining — powered by agentic LLM workflows.

Skills Demonstrated

NLP Agentic LLM Workflows RAG Vector Search (FAISS) Flask + SSE LangChain Prompt Engineering

Tools Used

Python Flask Phidata Mistral FAISS LangChain Bootstrap

Solution

Agentic pipeline orchestrates multiple LLM sub-tasks: NER, Sentiment Quintuples (Target, Feature, Sentiment, Opinion Holder, Time), Coreference Resolution, and Comparative Opinion Mining. Real-time outputs are streamed via Server-Sent Events, and users can download CSV/JSON results. RAG via FAISS improves grounding.

Approach

  1. Preprocess: Clean and normalize the review text.
  2. Retrieve Context (RAG): Use FAISS cosine similarity to fetch relevant examples (optional).
  3. Agent Pipeline: Run specialized agents — NER, Sentiment Quintuples, Coreference, Comparative.
  4. Aggregate: Merge agent outputs into clear, structured insights.
  5. Stream: Send results in real time via SSE for a responsive UX.
  6. Export: Download outputs as CSV or JSON. Batch mode supports CSV uploads and summaries.

Project Link(s)

Repository: aiml_project_doc

Single App Demo

Batch App Demo

Completed: 2025

Designed and Developed by Aradhya Pavan H S