Hi, I'm
Somesh Bagadiya

Developed during a hackathon, this AI journaling platform uses a custom agentic pipeline to store user information and respond conversationally like a friend. Built with Reflex for UI, FetchAI for the agent system, and implemented an OpenAI-powered RAG system with ChromaDB for contextual understanding of emotional patterns. Delivers personalized reflection prompts and empathetic responses to foster self-awareness and mental clarity.

Email intent classification pipeline that analyzes and categorizes email content by purpose and intent. Implemented a complete workflow from email preprocessing to intent classification using spaCy for NLP tasks and OpenAI GPT-3.5 Turbo with fine-tuned LLAMA 2 models. Incorporated custom NER for extracting key entities and sentiment analysis for content evaluation, achieving high classification accuracy across diverse email formats.

Developed a data collection framework for traffic optimization using GTAV as simulation environment. Implemented pipeline capturing video data, processing with YOLOv8 for vehicle detection, and generating time-series data on traffic density and signal status. Built a LSTM model using PyTorch to predict optimized signal timing based on vehicle counts, focusing on creating a reliable data generation system that bridges simulation and real-world traffic patterns.

Financial forecasting system processing Yahoo Finance data through Pandas for feature engineering and TensorFlow for model training. Developed two separate LSTM models: one for exact next-day value prediction (10% accuracy) and another for directional movement classification (74% accuracy). Implemented extensive data visualization of market trends and prediction results using Matplotlib and Seaborn.

Leveraged Dlib for face recognition and precise eye position tracking to implement three key features: focus/attention monitoring, cursor control using eyes / gaze tracking, and drowsiness detection with 90% accuracy. Developed custom algorithms to track eyeball movements in real-time using OpenCV and Python, creating a versatile system for both safety monitoring and hands-free computer interaction.

Freelance project for Artonifs developing a safety monitoring system for critical electronic systems measuring impedance and voltage readings in real-time. Implemented socket programming for hardware-PC communication with custom signal filtering. Developed rule-based algorithms using threshold detection, moving averages, and standard deviation calculations for statistical pattern recognition, achieving 98% accuracy in identifying electrical failures. Built monitoring dashboard with Tkinter. Successfully deployed to production with 2+ years of continuous operation.

Freelance project for Artonifs developing a PC interface and hardware connectivity solution for specialized communication systems. Implemented custom communication protocol with socket programming and serial connectivity for reliable data exchange in challenging environments. Built intuitive Tkinter-based GUI for system configuration and management, improving communication reliability by 70%. Successfully deployed to production environments with 3+ years of continuous operation.

Independent freelance project developing a full-featured Windows desktop application with installer for streamlining quotation processes. Built using Python and Tkinter with comprehensive product configuration, pricing tiers, and approval workflows. In production since Dec 2021, saving countless man-hours and significantly impacting business growth by expediting quote generation by 60%, directly contributing to new customer acquisition.

Autonomous vehicle prototype using Raspberry Pi (master) for computation and Arduino Uno (slave) for motor control via H-bridge. Implemented ADAS features including lane centering using Canny edge detection with Hough transform, collision avoidance with sonar sensors, and object detection using Haar Cascade classifiers in C++ OpenCV. Developed custom algorithms for real-time processing optimized for resource-constrained hardware across varied driving conditions.

Personal passion project leading a two-year research initiative where 200-300 volunteers answered laptop-related preference questions. Created custom web scrapers using Beautiful Soup, Selenium, and Python to build a comprehensive dataset of 1000+ laptops. Research focused on gathering user preferences and product specifications to inform the development of a personalized gadget recommendation system, which is currently in progress.

AR-enhanced virtual assistant evolved from an earlier voice assistant project. Built a processing pipeline that captures voice input, converts to text using Speech Recognition library, retrieves information via Google search with Beautiful Soup scraping, and delivers responses through text-to-speech. Enhanced with Unity 3D to add augmented reality visualization, face recognition, and multimodal interaction capabilities, creating an immersive assistant beyond traditional voice-only experiences.

Foundational project that later evolved into CREVA. Developed complete voice assistant pipeline: capturing audio input, converting speech to text using Speech Recognition library, retrieving information via Google search with Beautiful Soup web scraping, and delivering responses through text-to-speech synthesis.