Developed at TVS Sensing Solutions | Internship Project
Note: This is a portfolio project showcasing work completed during my Data Science internship at TVS Sensing Solutions.
At TVS Sensing Solutions, the business development team needed a tool to quickly generate competitor analysis for market positioning and strategic planning. The challenge was to create a customizable system that could:
- Generate competitor information based on specific product categories
- Focus on TVS's sensor manufacturing portfolio
- Provide structured, organized outputs for decision-making
- Deliver results faster than manual web research
The Challenge: Build a tool that generates product-specific competitor intelligence with automated data generation and visualization.
I built an AI-powered web application using Flask and Ollama Mistral LLM that generates competitor profiles based on user-specified criteria.
Input: User specifies product type (e.g., pressure sensors), market focus (Indian/Global), and number of competitors
Process: Dynamic Query Construction → Ollama Mistral LLM → Data Processing → Ranking
Output: Structured competitor profiles with interactive dashboard + Excel export
Designed intelligent prompt engineering that transforms user inputs into structured queries for the Ollama Mistral model. Implemented data cleaning and a ranking algorithm that sorts generated competitor data by market share and turnover.
| Metric | Before | After | Result |
|---|---|---|---|
| Analysis Tool | Manual web research | AI-powered generation | Faster results |
| Query Customization | Generic searches | Product-specific prompts | Targeted queries |
| Data Structure | Unorganized notes | Structured DataFrames | Analysis-ready |
| Output Format | Manual compilation | Dashboard + Excel export | Professional reports |
| Visualization | Static notes | 6+ Interactive charts | Better insights |
Real-World Results:
- ✅ Built AI-powered system using Ollama Mistral LLM for competitor data generation
- ✅ Created product-specific prompt engineering for TVS's sensor portfolio
- ✅ Delivered interactive dashboard with 6+ Plotly visualizations
- ✅ Generated structured Excel reports for business presentations
Architecture Overview:
┌─────────────────────────────────────────────────────────────────┐
│ USER INTERFACE │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │
│ │ Home Page │ │ Input Form │ │ Dashboard │ │
│ └──────────────┘ └──────────────┘ └──────────────┘ │
└────────────────────────────┬────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ FLASK WEB APPLICATION │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │
│ │ Routes │ │ Form Handler │ │ Template │ │
│ │ (/index) │ │ Validator │ │ Renderer │ │
│ └──────────────┘ └──────────────┘ └──────────────┘ │
└────────────────────────────┬────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ PROMPT CONSTRUCTION MODULE │
│ │
│ Input: Product Type, Market, Sub-segment, Technology │
│ Process: Combine → Format → Structure │
│ Output: "pressure sensors in automotive using MEMS" │
│ │
└────────────────────────────┬────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ DYNAMIC QUERY CONSTRUCTION │
│ │
│ Context: "I am an industrial expert in automotive..." │
│ Query: "List top 10 Indian companies manufacturing..." │
│ Format: JSON with 12 fields (Company, HQ, Market Share...) │
│ │
└────────────────────────────┬────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ OLLAMA MISTRAL LLM │
│ │
│ Model: mistral │
│ Input: Structured prompt with context │
│ Output: JSON array of competitor objects │
│ │
└────────────────────────────┬────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ DATA PROCESSING PIPELINE │
│ │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │
│ │ JSON Parser │→ │Data Cleaning │→ │Normalization │ │
│ └──────────────┘ └──────────────┘ └──────────────┘ │
│ │
│ • Remove markdown ```json``` tags │
│ • Convert "5%" → 5.0 │
│ • Convert "5 Billion USD" → 5000000000 │
│ • Extract "500-1000 employees" → 750 │
│ │
└────────────────────────────┬────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ RANKING ALGORITHM │
│ │
│ 1. Sort by Market Share (descending) │
│ 2. Then by Turnover (descending) │
│ 3. Select Top N competitors │
│ │
└────────────────────────────┬────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ OUTPUT GENERATION │
│ │
│ ┌──────────────────┐ ┌──────────────────┐ │
│ │ VISUALIZATIONS │ │ EXCEL EXPORT │ │
│ ├──────────────────┤ ├──────────────────┤ │
│ │ • Pie Chart │ │ • XlsxWriter │ │
│ │ • Bar Charts (3) │ │ • Formatted │ │
│ │ • Scatter Plot │ │ • All 12 fields │ │
│ │ • Histogram │ │ • Download ready │ │
│ └──────────────────┘ └──────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────────┘
Five-Layer Architecture:
Layer 1: User Input - Web form capturing product and market specifications
Layer 2: Flask Application - Backend processing and routing
Layer 3: AI Engine - Ollama Mistral for competitor data generation
Layer 4: Data Pipeline - Cleaning, structuring, and ranking
Layer 5: Visualization - Interactive Plotly dashboards and Excel export
┌──────────────────────────┐
│ USER │
└───────────┬──────────────┘
│
┌───────────────┼───────────────┐
│ │ │
▼ ▼ ▼
┌──────────────────┐ ┌─────────────┐ ┌─────────────┐
│ Input Product │ │View │ │Export Excel │
│ Specifications │ │Dashboard │ │Report │
└──────────────────┘ └─────────────┘ └─────────────┘
│
▼
┌──────────────────────┐
│ Generate Competitor │
│ Analysis │
└──────────────────────┘
│
▼
┌──────────────────────┐
│ Ask Custom Questions │
│ (Chatbot) │
└──────────────────────┘
Complete system capabilities showing user interactions
┌──────────┐
│ USER │
└─────┬────┘
│
▼
┌────────────────┐
│ Input Form │
│ • Product Type │
│ • Market │
│ • Count │
└────────┬───────┘
│
▼
┌────────────────┐
│ Build Prompt │
│ Construction │
└────────┬───────┘
│
▼
┌────────────────┐
│ Ollama Mistral │
│ LLM │
└────────┬───────┘
│
▼
┌────────────────┐
│ Parse JSON │
│ Response │
└────────┬───────┘
│
▼
┌────────────────┐
│ Clean & Norm. │
│ Data │
└────────┬───────┘
│
▼
┌────────────────┐
│ Rank by Market │
│ Share/Turnover │
└────────┬───────┘
│
┌─────────┴─────────┐
▼ ▼
┌─────────────┐ ┌─────────────┐
│ Plotly │ │ Excel │
│ Dashboard │ │ Export │
└──────┬──────┘ └──────┬──────┘
│ │
└─────────┬─────────┘
│
▼
┌──────────┐
│ USER │
└──────────┘
End-to-end processing pipeline from input to output
Landing page with navigation to analysis modules
User specifies market type, product segment, and competitor filters
Additional filters for sub-segment and technology specifications
Indian/Global market toggle and instant Excel download functionality
Interactive dashboard with multiple visualization types
Pie chart showing competitive landscape distribution
Bar chart comparing competitor market positions
Organizational scale comparison across competitors
Scatter plot identifying company size and market position
Conversational interface for ad-hoc queries
Structured data export for reports
1. User Input Processing Module
- Flask form handling for product specifications
- Market type selection (Indian/Global)
- Competitor count configuration
- Dynamic field validation
2. Prompt Construction Engine
- Product description builder
- Sub-segment combination logic
- Technology specification integration
- Structured query formation
3. AI Integration Layer
- Ollama Mistral LLM integration
- Context-aware prompt engineering
- JSON response parsing
- Error handling for invalid responses
4. Data Processing Pipeline
- Market share normalization (%, Billion, Million, USD)
- Turnover standardization across currencies
- Employee range extraction (e.g., "500-1000" → 750)
- Data type conversion and cleaning
5. Ranking Algorithm
- Dual-criteria sorting (Market Share × Turnover)
- Top N competitor selection
- Handles missing values
6. Visualization Module
- 6 distinct Plotly chart types
- Market share pie charts
- Employee count bar charts
- Efficiency scatter plots
- Revenue distribution histograms
- Interactive HTML rendering
7. Export Service
- XlsxWriter integration
- Structured Excel formatting
- In-memory file generation
- Browser download handling
8. Web Interface
- Flask routing and template rendering
- Responsive Bootstrap UI
- Real-time dashboard updates
- Custom prompt chatbot interface
| Category | Technologies | Purpose |
|---|---|---|
| Backend | Python 3.11, Flask 3.0.0 | Web framework, routing, business logic |
| AI/NLP | Ollama 0.6.2, Mistral LLM | Competitor data generation |
| Data Processing | Pandas 2.1.1 | Data manipulation, cleaning, ranking |
| Visualization | Plotly 5.17.0 | Interactive charts and dashboards |
| Export | XlsxWriter | Excel report generation |
| Frontend | HTML5, CSS3, Bootstrap | Responsive UI |
| Testing | Pytest | Unit testing |
✅ Dynamic Form Processing - Captures user specifications for competitor analysis
✅ LLM Integration - Uses Ollama Mistral to generate structured competitor data
✅ Smart Data Cleaning - Normalizes multiple data formats from LLM responses
✅ Ranking Algorithm - Sorts competitors by market share and turnover
✅ Interactive Dashboards - 6 visualizations for data analysis
✅ Excel Export - Professional reports ready for download
✅ Customizable Queries - Market type, product segment, sub-segment, technology filters
✅ Chatbot Interface - Ad-hoc queries with conversational AI
- Large Language Model (LLM) integration
- Prompt engineering for structured outputs
- JSON parsing and validation
- Context-aware query construction
- Flask web application development
- Dynamic form handling
- File generation and download
- Error handling
- Data cleaning pipeline
- Multi-format normalization
- Currency and percentage standardization
- Ranking algorithm implementation
- Dashboard design
- Interactive Plotly charts
- Multiple visualization types
- Export functionality
- Frontend UI/UX design
- Backend logic implementation
- Template rendering
- End-to-end system integration
1. Requirements Gathering - Understood competitor analysis needs from TVS team
2. System Design - Designed Flask application with LLM integration
3. Prompt Engineering - Built dynamic query construction from user inputs
4. AI Integration - Implemented Ollama Mistral with structured output validation
5. Data Pipeline - Developed cleaning and ranking algorithms
6. Visualization - Created 6 interactive dashboards
7. Testing - Built pytest test suite for validation
Created test suite using pytest to validate:
- Page loading and routing
- Custom prompt functionality
- Response accuracy against expected outputs
# Sample test structure
def test_index_page(client):
response = client.get("/")
assert response.status_code == 200
def test_response_accuracy():
expected_output = ["Bosch", "TE Connectivity", "Sensata"]
# Validate chatbot responsesI'm a Data Analytics Engineering graduate student at Northeastern University seeking co-op/full-time Data Analyst or Data Scientist roles.
This project demonstrates my ability to:
- ✅ Build AI-powered applications using Large Language Models
- ✅ Design interactive dashboards for data visualization
- ✅ Develop end-to-end web applications
- ✅ Work independently on complete project lifecycle
Interested in discussing this project?
📧 Email: varaalakshime.l@northeastern.edu
💼 LinkedIn: [https://www.linkedin.com/in/varaalakshime-v]
Available for Co-op: May 2025 - December 2025
Developed during Data Scientist internship at TVS Sensing Solutions Private Limited, Madurai - A leading manufacturer of advanced sensing technologies and electronic components under the TVS Sensing Solution Group, specializing in pressure sensors, speed sensors, temperature sensors, and position sensors for automotive and industrial applications.
- Ollama Documentation - LLM integration and prompt engineering
- Flask Documentation - Web framework implementation
- Plotly Documentation - Interactive visualizations
- Pandas Documentation - Data manipulation and analysis
- TVS Sensing Solutions Private Limited - Internship opportunity and domain expertise
- Dr. M. S. Sabitha (External Guide) - Technical guidance and project mentorship
- Dr. C. Mahadevi (Internal Guide) - Academic supervision