Prafful

Tractor Valuation System

• Built AI pipelines to analyze tractor images and RC documents using YOLO, OpenCV, and OCR tools.
• Implemented models to check tyre condition, rust, number plates, and calculate tractor value using depreciation and price-matching logic.
• Integrated with external APIs to fetch tractor registration details and verify ownership.

Multilingual Product Search Assistant

• Developed a WhatsApp chatbot that understands farmer queries in Hindi, Hinglish, and English.
• Created structured datasets and trained models to match user queries with correct agri products.
• Used vector search (FAISS) and LLMs to improve product recommendations.

Automation & Data Handling

• Designed pipelines to process farmer messages, generate structured outputs (JSON), and send replies automatically.
• Stored all interactions in a database with proper logs for tracking and debugging.
• Worked on cost optimization and speed improvement for AI model usage.

Team Contribution

• Designed clear prompts and workflows for AI assistants.
• Worked with frontend and backend developers to integrate AI models into production.
• Handled end-to-end deployment on servers with Python + Flask APIs.

Automated Telecom Surveying Tool

• Developed Image Detection and Image Classification to detect and classify between Pit or Manhole
• Developed Image Detection and Classification to detect and classify the type of Pit (i.e. Open, closed, curved, Edged, etc)
• Developed an algorithm based on OpenCV to detect a specific-colored wire inside the pit and extract it.
• Developed an algorithm based on OpenCV to detected the pit surrounding and classify between three classes between (i.e. Grass, Cement and Muddy).
• Developed a Yolo based model to detect crack on Pit and it's level of damage (cracked, broken or caved in).

Automatic Easement Polygons Creation using AI/ML on AutoCAD

• Automated the process of creating easement polygons, previously manual
• Extracted data from Ordinance Survey PDF forms; converted PDFs into images; applied OCR processing
• Extracted required parameters using heuristic and text classification algorithms.
• Integrated data with an AutoCAD script (LISP) that draws the Polygons as required , reducing time by 60% and increasing profits.

New world Satellite imagery Feature extraction

• Leveraged YOLO and Unet for feature extraction from satellite images, facilitating tasks such as urban planning, change detection and environmental monitoring.
• Trained and extracted different features from the satellite images (i.e. Bus Lane, cycle lane, road traffic markings etc)
• Developed a solution to divide large images into small pieces, process it and then deliver it after merging so as to easing and fastening the processing

Her majesty’s Land Record (HMLR) : Text Feature Extraction

• Created a technique which uses both EasyOCR and PyTesseract to extract feature from typed documents
• Developed a filtering algorithm to filter the extracted text and save the desired data to database.
• In case of handwritten text, used Google OCR and extracted the feature and filtered the text and saved it to database.