Joint Squad
FlavourLens
"Code, Solve realistic problems, Inspire generations"
Submitted
May 22, 2026, 9:56 AM
Last Updated
May 23, 2026, 6:25 AM
Project Links
Problem Statement
Restaurants today struggle to truly understand changing customer preferences and market dynamics. Most feedback is scattered across review platforms, social media, and offline conversations, making it difficult for restaurant owners to identify what customers actually like or dislike about their food, service, ambiance, or staff. Existing review systems provide only generic ratings and lack deep, actionable insights tailored specifically for restaurant growth. FlavourLens solves this problem by providing a dedicated AI-powered platform exclusively for restaurants. The system analyzes customer reviews using Aspect-Based Sentiment Analysis (ABSA), identifies detailed sentiments about different aspects of the dining experience, and transforms raw feedback into meaningful business insights. By combining real-time analytics, knowledge graphs, and an AI assistant, FlavourLens helps restaurants better understand customer needs, adapt to market trends, improve service quality, and make smarte
Solution & AI Usage
We built FlavourLens as a dedicated AI-powered feedback intelligence platform for restaurants using a FastAPI backend and a Flutter frontend. Customer reviews are processed through a lightweight Aspect-Based Sentiment Analysis (ABSA) pipeline that identifies sentiments related to food, service, ambiance, and staff. For AI, we used fine-tuned DistilBERT models in PyTorch because they are fast, efficient, and suitable for real-time inference on limited GPU resources. A secondary escalation layer uses Groq’s LLaMA 3.3 70B API only for complex reviews involving sarcasm, contradictions, or low-confidence predictions, ensuring both speed and accuracy. The processed insights are stored in Firebase Firestore for dashboard visualization and Neo4j AuraDB for relationship-based analysis and chatbot context generation. This combination allows restaurant owners to interact with their business insights through a grounded AI assistant powered by their own customer data.
Full Description
FlavourLens — AI-Powered Restaurant Intelligence Platform Overview
FlavourLens is a dedicated AI-powered platform built specifically for restaurants to help them better understand customer feedback, market dynamics, and changing consumer preferences. Traditional review platforms only provide star ratings and scattered comments, making it difficult for restaurant owners to extract meaningful insights. FlavourLens transforms raw customer reviews into actionable business intelligence.
The platform uses Aspect-Based Sentiment Analysis (ABSA) to analyze reviews at a granular level and identify customer opinions about specific aspects such as food, service, ambiance, and staff.
Key Features
- AI-Powered Review Analysis Extracts important aspects from customer reviews Detects sentiment for each aspect individually Supports Positive, Negative, and Neutral sentiment classification Handles sarcasm, negation, and contextual contradictions
- Smart Escalation System Fast local inference using lightweight transformer models Automatically escalates difficult reviews to a cloud LLM only when needed Improves accuracy while reducing API cost and latency
- Restaurant Knowledge Graph Stores structured restaurant insights inside Neo4j AuraDB Connects Restaurants, Reviews, Tables, Dates, and Aspects through relationships Enables advanced querying and trend discovery
- AI Chat Assistant for Restaurant Owners
Owners can ask questions like:
“What are customers complaining about most?” “Which table receives the worst feedback?” “How is food sentiment trending this week?”
The chatbot generates responses grounded entirely on restaurant data from Neo4j.
- Real-Time Dashboard Review summaries Sentiment breakdowns Trend monitoring Complaint tracking Performance insights Architecture Frontend Built using Flutter Cross-platform UI for restaurant management Backend FastAPI server running on Kaggle GPU environment Uvicorn used as ASGI server ngrok used for secure public HTTPS tunneling AI Pipeline Local Models Fine-tuned DistilBERT Aspect Extractor DistilBERT Sentiment Classifier Implemented using PyTorch Cloud AI Groq API with LLaMA 3.3 70B Used only for low-confidence or complex reviews Database Layer Firebase Firestore
Stores:
Review documents Dashboard data User-facing analytics Neo4j AuraDB
Stores:
Restaurant knowledge graph Relationship-driven insights Chatbot context Workflow Customer submits review from Flutter app Review reaches FastAPI /submit endpoint Text preprocessing detects: emojis sarcasm negation patterns Local DistilBERT models perform ABSA Complex reviews escalate to Groq LLaMA Final insights stored in Firestore and Neo4j Restaurant owners interact with AI chatbot through /chat Challenges Faced
- Handling Sarcasm & Contradictions
Customer reviews often contain indirect sentiment like:
“Amazing food… waited 2 hours.”
This required additional preprocessing and escalation logic.
- Balancing Speed and Accuracy
Running large models on every rev
Tech Stack
Screenshots
5 attached
