Airline Review Retrieval-Augmented Generation System

Built a retrieval-augmented generation pipeline over airline customer reviews to surface actionable insight from unstructured feedback.

PythonMongoDB AtlasLangChainLangSmithStreamlit

Retrieval-augmented generation pipeline

01Customer reviews

02Vector index (Atlas)

03Semantic retrieval

04LangChain + prompt

05Streamlit answer

No labelled benchmark exists for this academic implementation, so results were assessed qualitatively rather than scored, shown here as the pipeline architecture instead of a fabricated accuracy metric.

Problem / Research Question

Airline customer reviews contain rich but unstructured feedback. Traditional keyword search struggles to answer open-ended questions about recurring themes; the goal was context-aware, retrieval-grounded answers instead.

Dataset

Airline customer review text data.

Methodology

01Indexed review text using MongoDB Atlas Vector Search for semantic retrieval.
02Built a RAG pipeline with LangChain, applying prompt engineering for context-aware responses.
03Used LangSmith to trace and evaluate the pipeline during development.
04Deployed the system via a Streamlit interface.

Key Findings

Semantic vector retrieval combined with prompt engineering allowed the system to answer open-ended questions about review content rather than only exact keyword matches.

Limitations

As an academic RAG implementation, the system wasn't evaluated against a labelled benchmark for answer accuracy; results were assessed qualitatively during development.