thevirtualhero/business-analyzer

Purpose

Business-Analyzer is an LLM-assisted review-intelligence project for local businesses. It turns Google Maps-style customer reviews into sentiment, themes, and business recommendations. The repository is a compact prototype rather than a clearly separated production application.

Entry Points

• google maps review analyser.ipynb is the likely exploratory/interactive analysis entry point for loading and analyzing reviews.

• llm and code.py is the likely scripted LLM integration or automation entry point.

• README.md describes intended Streamlit/Gradio deployment, but no frontend source file is present in the tree.

Review Data Boundary

• ReviewsDetails.csv is the repository’s local review-data input.

• The intended upstream acquisition boundary is SerpAPI/Google Maps Reviews, as described in README.md; no API client or credentials file is visible.

• The CSV/notebook boundary separates raw or simulated review records from NLP and LLM analysis.

NLP and Classification Stage

The analysis layer is intended to clean review text and derive structured signals:

• Sentiment classification distinguishes positive, negative, and complaint-oriented feedback.

• Thematic categorization identifies recurring operational topics such as service speed, pricing, and staff behavior.

• This stage is represented by the workflow-oriented artifacts google maps review analyser.ipynb and llm and code.py, with the dataset in ReviewsDetails.csv.

LLM Insight Generation Stage

The core reasoning flow uses prompt-driven LLM processing to convert review-level signals into summaries and recommendations. The stated model ecosystem is Hugging Face and Langflow (README.md), likely invoked or prototyped through llm and code.py. Outputs are intended to include operational suggestions, trend-based recommendations, and competitor-related highlights.

Presentation and Delivery Boundary

The README positions Streamlit or Gradio as the user-facing deployment runtime, but the repository contains no corresponding app module. Therefore, the current visible delivery mechanism is the notebook (google maps review analyser.ipynb), while a dashboard layer remains an intended integration boundary rather than a confirmed implemented component.

Runtime and External Dependencies

Material runtimes/services named by the project are Python, notebook execution, Hugging Face models, Langflow, and SerpAPI. README.md is the only visible dependency/design declaration; there are no package manifests, deployment descriptors, Langflow exports, or frontend source files in the provided tree. Engineers should treat model selection, API ingestion, and dashboard serving as integration points needing verification inside llm and code.py and the notebook.