In today’s dynamic online communities, users frequently seek clarity on enterprise technologies, frameworks, and tools. However, many relevant posts go unanswered or receive inconsistent feedback. To address this gap, LimeLight was developed—an intelligent assistant designed to autonomously detect relevant discussions, retrieve contextual data, and generate high-quality, sentiment-aware responses in Niche communities like Reddit and Slack.

This project demonstrates how Retrieval-Augmented Generation (RAG), modern language models, and sentiment analysis can be combined to enrich community interactions in a scalable and meaningful way.

Project Overview

LimeLight is a modular, AI-driven system integrated into a community platform. It automatically identifies posts related to enterprise technologies, retrieves relevant information from a vector database, and responds with well-formed, context-sensitive comments. It also considers the sentiment of user posts to adjust tone and engagement accordingly.

Key Features

1. Intelligent Content Detection

LimeLight identifies relevant content using a hybrid method of:

  • Keyword filtering (e.g., references to enterprise platforms, services, or tools)
  • Semantic similarity via sentence embeddings

This ensures the assistant only engages with content aligned to its knowledge domain.

2. RAG-Based Contextual Responses

When a relevant post is detected:

  • Associated data is retrieved from a ChromaDB vector store
  • A LangChain RAG pipeline feeds this context into LLaMA 3, running locally via Ollama
  • A response is generated and inserted into the conversation thread

3. Sentiment-Aware Interaction

LimeLight evaluates the sentiment of the original post (positive, neutral, or negative) and adjusts its tone accordingly. For example:

  • A negative sentiment triggers more empathetic or supportive language
  • Neutral or factual posts receive concise and informative replies

4. Automated Post Engagement

On post creation, the system performs the following:

  • Checks if the content is within scope
  • If relevant, triggers generate_response() in bot.py
  • Inserts a comment into the database, visible in the discussion thread
  • Tags the response as authored by "LimeLight" with a timestamp

Technology Stack

Component Technology
Backend FastAPI
Frontend HTML, CSS, Jinja2
Database MongoDB (accessed via Motor)
LLM LLaMA 3 (served via Ollama)
RAG Pipeline LangChain + HuggingFace
Vector Store ChromaDB
Embeddings Sentence Transformers
ML Extension Sentiment Analysis (Planned)
Dashboard HighCharts

Installation & Usage

Clone the repository:

git clone https://github.com/VaishnaviUnnikrishnan/NicheBot.git
cd NicheBot

Set up the virtual environment:

python -m venv venv
source venv/bin/activate       # For Mac/Linux
venv\Scripts\activate          # For Windows

Install dependencies:

pip install -r requirements.txt

Start the application:

uvicorn main:app --reload

Ensure LLaMA 3 is active using Ollama:

ollama run llama3

Visit the platform at http://localhost:8000

System Architecture 

User Post
   │
   ├──> Relevance Detection (keywords + embeddings)
   │
   ├──> ChromaDB Vector Retrieval
   │
   ├──> LangChain RAG Pipeline
   │
   ├──> LLaMA 3 Generates Response
   │
   ├──> Sentiment Analysis
   │
   └──> Auto-Comment as "LimeLight"

The architecture is modular, with separation of concerns across:

  • main.py – routing and server logic
  • response.py – relevance and sentiment processing
  • bot.py – RAG pipeline and language generation

Future Enhancements

  • Replace keyword detection with an LLM-based classifier for improved topic relevance
  • Integrate advanced sentiment classification using transformer-based models
  • Expand to support other domains beyond enterprise tech

Sneakpeak

Interactions in Community

Limelight's response

User Feedback on the responses given by Limelight

Contact

For queries, feedback, or collaboration opportunities, contact:

Email: v.ukrishnan8@gmail.com