ByteFlow Mart · Sellers have hit a competitive intelligence wall where the sheer velocity of cross-platform pricing shifts, sentiment nuances in thousands of reviews, and shifting competitor tactics has outpaced human analytical capacity. Traditional data scraping fails because it lacks the autonomous reasoning to connect disparate signals—like a specific material complaint in a reviewto a pricing drop in a rival'scatalog. This creates a strategic blind spot that requiresan agentic layerto independently monitor,synthesize, and proactively suggest pivot strategies before market opportunities evaporate.
PriceRadar is a local-first, privacy-preserving, agentic competitive intelligence platform designed for Indian e-commerce sellers.
It autonomously monitors competitor listings across major marketplaces, analyzes pricing gaps, delivery advantages, discounts, and review sentiment, and generates priority-ranked, one-click actionable strategies using a hybrid rule engine and on-device Large Language Model (LLM).
🔒 No seller data leaves the machine.
⚡ No external API dependency.
🧠 Real-time strategic intelligence.
- Problem Statement
- What Is PriceRadar?
- Full Workflow
- System Architecture
- Why This Approach
- LLM Selection — Qwen3-8B-Instruct
- How the LLM Is Used
- Tech Stack
- Project Structure
- Local Setup
- Docker Deployment
- API Reference
- Dataset
- License
Sellers have hit a competitive intelligence wall where the velocity of cross-platform price changes, delivery competition, discount wars, and sentiment signals has exceeded human analytical capacity.
Traditional scraping tools provide raw data but lack reasoning ability to connect signals — for example:
- Price drops triggered by negative reviews
- Faster delivery influencing conversion rates
- Discount battles across marketplaces
- Feature complaints impacting demand
This creates a strategic blind spot requiring an agentic layer that continuously monitors, synthesizes, and proactively recommends actions before revenue loss occurs.
PriceRadar is a full-stack competitive intelligence system that:
✔ Scrapes live competitor data
✔ Matches equivalent products across platforms
✔ Computes market gaps and trends
✔ Analyzes review sentiment
✔ Generates actionable strategies
✔ Applies optimizations with one click
✔ Operates fully offline with local AI
- Flipkart
- Amazon
- Croma
- Reliance Digital
- Vijay Sales
Seller Input ↓ Automated Scraping ↓ Product Matching ↓ Market Analysis ↓ Strategy Generation (Rule Engine) ↓ LLM Strategic Advisory ↓ Seller Decision / One-Click Apply ↓ Continuous Monitoring ↓ Alerts & Re-Optimization
- Optimal pricing recommendations
- Delivery improvements
- Promotion strategies
- Competitive positioning insights
- Automated update suggestions
Frontend → FastAPI Backend → Strategy Engine → LLM Layer → Data Storage → Scraper → Scheduler
Rule Engine
Deterministic logic for instant, reliable actions.
LLM Layer
Context-aware reasoning and strategic insight.
Scheduler
Autonomous 24-hour monitoring loop.
Data Layer
Lightweight JSON storage for portability.
Pure LLM systems are slow and unpredictable.
Pure rule systems cannot handle novel situations.
PriceRadar combines both:
| Capability | Rule Engine | LLM | Hybrid |
|---|---|---|---|
| Speed | ⚡ Instant | Slow | Fast |
| Reliability | High | Variable | High |
| Reasoning | Low | High | High |
| Offline Capability | Yes | No | Yes |
| Explainability | High | Medium | High |
Result: Deterministic core + intelligent advisory layer
Chosen for:
- Excellent instruction adherence
- Strong numerical reasoning
- CPU-friendly quantized deployment
- Fully offline inference
- Apache 2.0 license
- No external API requirement
Qwen3-8B-Instruct Q4-K-M (Quantized)
bash ollama pull qcwind/qwen3-8b-instruct-Q4-K-M ollama serve
The LLM acts as a strategic advisor — not the decision engine.
- Explaining strategy recommendations
- Interpreting review sentiment
- Suggesting competitive positioning
- Providing marketing insights
- Highlighting risks and opportunities
All operational actions remain deterministic for reliability, ensuring consistent system behavior even if the LLM is unavailable.
- HTML / CSS / JavaScript
- Streamlit Dashboard
- FastAPI
- Uvicorn
- Python 3.11+
- Qwen3-8B-Instruct (Q4-K-M quantized model)
- Ollama — local LLM runtime
- Playwright — headless browser scraping
- Chromium Engine — executes JavaScript and waits for DOM
- Extraction from JS-rendered pages
- Flipkart
- Amazon
- Croma
- Reliance Digital
- Vijay Sales
schedule— 24-hour autonomous monitoring loop- Alert system for price and discount changes
Lightweight JSON flat files for portability and zero external dependencies.
- Products
- Orders
- Users
- Competitor data
- Bundles
- Seller profile
project/ ├── ai/ # LLM modules ├── strategy/ # Rule engine & apply logic ├── scraper/ # Competitor scraping ├── scheduler/ # Autonomous monitoring ├── data/ # JSON datasets ├── dashboard.py # Strategy dashboard ├── server.py # FastAPI backend ├── requirements.txt
- Python 3.11+
- Ollama installed
bash git clone cd priceradar pip install -r requirements.txt pip install playwright playwright install chromium
- Python 3.11+
- Ollama installed
bash git clone cd priceradar pip install -r requirements.txt pip install playwright playwright install chromium
- Python 3.11+
- Ollama installed
bash git clone cd priceradar pip install -r requirements.txt pip install playwright playwright install chromium
🐳 Docker Deployment Development Mode docker compose --profile dev up --build
📡 API Reference Seller APIs Endpoint Description GET /api/seller/products List products POST /api/seller/analyze Analyze competition POST /api/seller/alerts/scan-all Run monitoring POST /api/seller/alerts/ai-strategy/{pid} Generate AI insights Strategy Dashboard APIs Endpoint Description GET /api/strategy/data Market dataset POST /api/strategy/apply-all Apply optimizations POST /api/strategy/reset Restore original data
Sam Naveenkumar V
Total: 250 real review samples
📄 License
CIT License — free for academic and commercial use.