Agentic AI · M.Tech Dissertation

Enhanced Feedback Analysis

My BITS Pilani M.Tech dissertation: an agentic pipeline that turns unstructured open-text feedback into structured, actionable insight at scale.

Live Demo All Projects

Technical Information

Project Type
Agentic AIInformation ExtractionFeedback Analytics
Tech Stack
PythonFlaskLangChainFAISSSentence TransformersSLMFew-Shot LearningPrompt EngineeringSSE
Topics / Modules
Coreference ResolutionNER AgentQuintuple ExtractionComparative InsightsBatch Processing
Links
Live Demo

Project Summary

An agentic system that analyses large volumes of open-text feedback. Using few-shot prompting, LangChain agents and FAISS-backed retrieval, it extracts entities, resolves coreferences and surfaces comparative insights – with batch processing and live streaming of results.

Skills Demonstrated

Few-Shot PromptingAgentic OrchestrationCoreference ResolutionNamed Entity RecognitionSemantic RetrievalBatch Pipelines

Solution

A multi-agent pipeline (NER, coreference and quintuple-extraction agents) processes feedback in batches, stores embeddings in FAISS for retrieval, and streams progress to the UI via Server-Sent Events – producing downloadable structured JSON.

Approach

  1. Define: Frame feedback analysis as structured extraction (entities, relations, sentiment).
  2. Agents: Compose NER, coreference and quintuple-extraction agents with LangChain.
  3. Retrieve: Embed text with Sentence Transformers and index in FAISS.
  4. Process: Run single and batch jobs, streaming progress over SSE.
  5. Compare: Generate comparative insights and summaries across feedback.
  6. Export: Let users download structured JSON of the results.