Completed projects
Completed projects
I build practical AI solutions that solve real problems. My work spans computer vision, LLM powered automation and data driven systems used by clients across different industries.
Embedded Files
Document AI • OCR • RAG • LLM Pipelines
This project automates large-scale document processing, from file collection to structured data extraction. An advanced OCR engine converts PDFs, Word files, and images into clean, reliable datasets. A RAG intelligence layer lets users search, summarize, and analyze thousands of documents with high accuracy, creating a fast, scalable, and fully automated document-analysis workflow.
Computer Vision • YOLO Training • Automation
AI system for detecting and classifying structural defects from facade images. Includes a complete pipeline for data collection, model training, desktop app development and automated Word reporting. Provided a Windows application to upload photos, review detections and export structured reports.
OCR • TensorFlow Lite • On-Device AI • Receipt Parsing
A custom offline OCR system designed to process receipt images directly on Android devices without any API or cloud service. The engine is built from a tailored preprocessing pipeline, a fine-tuned OCR model trained on the client’s real receipts, and a lightweight LLM component for structured field extraction. The solution runs entirely on-device, optimized for mobile constraints while preserving strong accuracy across multiple receipt formats and lighting conditions.
Computer Vision · Floor Plans · OpenCV · YOLOv10
A desktop application that analyzes 2D floor plans and automatically calculates room areas, usable wall lengths, perimeters and volumes with high accuracy. The system removes doors and windows from measurements, converts pixel geometry into real-world units and exports both JSON and Excel files ready for professional use.
OCR • LLM • GPT • Data Extraction • Automation
A lightweight AI prototype that extracts structured data from PDF documents using advanced OCR and LLM reasoning. The system reads invoices and technical documents, identifies key fields with high accuracy and exports everything into a clean, ready-to-use Excel file. Built in Python with GPT and custom OCR logic, optimized for real-world document variability.
Data Automation · Lead Intelligence · Multi-Source Collection
Across multiple client projects, I built automated systems that collect business data from diverse online sources and convert it into structured, high-quality leads. These solutions extract names, emails, company details and metadata, enrich them with AI classifiers and validate them in real time. Each automation adapts to the target source, remaining stable even when layouts change.
Computer Vision • OCR • YOLOv11 • Automation Pipeline
This project delivers an end-to-end solution for extracting operational data from thousands of train-display images captured every 1–3 seconds during a 3-hour ride. The system combines OCR, YOLOv11 detection models and pixel-level measurement logic to read energy metrics, traction forces, current, and speed even under difficult conditions such as motion blur, reflections, tunnel darkness and tilted camera angles.
Solana • Trading Bot • Real-Time Alerts • Crypto AI
This project delivers a complete real-time intelligence and trading system for the Solana blockchain. The application continuously monitors on-chain activity, identifies synchronized token purchases across multiple wallets and reacts immediately by sending Telegram notifications and executing trades based on predefined rules.
Computer Vision • Perspective Correction • Warping
AI tool that automatically detects postcards in raw photos and applies perspective correction and neural warping to “unfold” bent, skewed or curved edges without cropping any part of the card. Handles blur, tilt, reflections and extreme lighting, turning crumpled postcard photos into clean, rectangular images ready for your existing Photoshop and upscaling workflow.
OCR • NLP • Data Parsing • Automation
A custom Python system that digitizes the 1922 Zingarelli dictionary starting from TIFF scans and converts it into structured JSONL files. The pipeline restores degraded pages, performs high-accuracy OCR and parses lemmas, meanings, derivatives and examples into clean, machine-readable records. Built as a standalone Python tool with full documentation, ready to be used for research, linguistics and archive preservation.
Page updated
Google Sites
Report abuse