This portfolio showcases a selection of diverse and innovative projects:

Technical Domains: NLP | Computer Vision | Anomaly Detection | Predictive Analytics | Recommendation Systems | Reinforcement Learning | Healthcare AI | FinTech | E-commerc...

Covering the complete development cycle: model architecture creation, custom training, fine-tuning, and production deployment. Each project demonstrates deep technical expertise in solving real-world industrial challenges,
spanning natural language processing and computer vision to predictive analytics and reinforcement learning. This end-to-end approach illustrates complete mastery of AI model development,
from algorithmic design to specialized industrial solutions.

Interactive AI Projects Section

Section dedicated to interactive AI projects developed using an advanced integration approach. These applications embed their user interfaces directly into the portfolio architecture, creating real-time functional demonstrations that coexist with other project categories. Complete portfolio source code access is available here {link}, enabling in-depth analysis of technical implementations, API communication architectures, and modular integration strategies.
This methodology demonstrates my comprehensive technical mastery of full-stack development in artificial intelligence.

AI Customer Support

Highlights

• Real-time voice assistant handling phone calls via Twilio with natural speech interaction
• RAG-powered answers using LangChain and MiniLM for accurate, knowledge-based responses
• Dynamic response generation with OpenAI GPT and text-to-speech conversion
• Adaptive wait sounds that match processing time for premium user experience
• Comprehensive call session management with status tracking and metrics
• Audio caching system for fast repeated answers to common questions

Tags: Conversational AI, customer support, NLP, Twilio, RAG, LangChain, MiniLM

Try It

GitHub Repo

Read More

E-Commerce Assistant

Highlights

• Advanced conversational platform with transactional context management
• Modular architecture with 3 specialized agents (products, orders, support)
• Real-time processing with < 2s latency through prompt optimization
• RAG integration with ChromaDB knowledge base
• 98.7% intent classification accuracy on 250K test queries
• Persistent session mechanism with encrypted entity tracking

Tags: FastAPI, LangChain, ChromaDB, WebSocket, GPT-4

Try It

GitHub Repo

Read More

AI Data Analyst

Highlights

• Autonomous data processing pipeline powered by GPT-4o for dynamic analysis of CSV, Excel, and text datasets
• Smart code generation with validation and auto-correction capabilities for reliable data transformations
• Real-time progress tracking with WebSocket communication for enhanced user experience
• Interactive visualization generation with matplotlib for immediate insights from complex datasets

Tags: Data Analysis, Machine Learning, Python, Pandas, Matplotlib, GPT-4, FastAPI

Try It

GitHub Repo

Read More

AI Research

Highlights

• Perplexity-Like with real-time web research, intelligent source scraping and content synthesis
• Dual-mode system: Standard research + Logical reasoning from knowledge base
• WebSocket integration for live progress tracking and status updates
• Automatic fact extraction and storage with FAISS vector database
• GPT API integration for advanced text generation and analysis
• Self-correcting pipeline with error handling and quality validation

Tags: Web Scraping, FastAPI, FAISS, Knowledge Base, WebSocket, GPT-4, LangChain

Try It

GitHub Repo

Read More

Movie Recommendation

Highlights

• Intelligent API for personalized movie recommendations based on user-selected titles
• Data processing pipeline that cleans, merges, and enriches TMDB movie datasets
• Similarity computation using Scikit-Learn (CountVectorizer + cosine_similarity)
• Dynamic movie poster fetching from TMDB API for enhanced user experience
• Modern FastAPI backend with efficient REST endpoints
• Genre, cast, and director-based recommendation algorithm

Tags: Recommendation System, TMDB API, FastAPI, Scikit-Learn, Data Processing

Try It

GitHub Repo

Read More

OCR Translation

Highlights

• Multi-format image support (PNG, JPG) with accurate text extraction using PaddleOCR
• Automatic language detection and instant translation for 6 languages
• Interactive preview with zoom functionality to examine detected text areas
• FastAPI backend with real-time processing status updates
• Integration with DeepL for high-quality translations
• Visual overlay of detected text boxes with original/translated text comparison

Tags: OCR, Image Processing, Translation, FastAPI, PaddleOCR, DeepL

Try It

GitHub Repo

Read More

Categorized Project Portfolio

Want to explore my projects by category? Below, you'll find my work organized by application domain for easy navigation.
Each section groups specialized projects within a particular technical area. Feel free to check out my GitHub for detailed implementations and technical documentation.

Natural Language Processing

Text Generation with GPT2 Fine-Tuned PDF Text Summarization Automatic Translation with Seq2Seq User-Feedback Analysis AI Customer Support E-Commerce Assistant OCR Transcription

Computer Vision

Medical Image Segmentation Object Recognition in Real-Time Videos Improved Image Quality Vision-Language Models Super-Resolution Medical Images with ESRGAN Stable Diffusion Fine-Tuned Natural Style Transfer Liveness Detection & Anti-Spoofing Vision-Language Models AI Agent Executor

Data Analysis & Prediction

Demand Prediction and Inventory Optimization Churn Analysis for Subscriptions Optimization of Algorithms Trading Strategies AI Data Analyst

Fraud & Anomaly Detection

Fraud Detection Anomaly Detection in Network Traffic

Recommendation & Personalization

Learning-to-Rank for E-commerce Movies Recommender

Generative AI

10M Parameters Transformer (MoE) AI-Powered Map LLMs Dataset Generator

Foundational Machine Learning: Neural Network From Scratch

Built a full neural network from zero - no libraries, not even numpy & pandas. Every function, metric, and update step was hand-coded in pure Python.

• Manual implementation of forward/backward pass, gradients, activation, loss, and optimizer
• All math (matrix ops, derivatives) written out, line by line
• Metrics and training loop coded from first principles

Why? To prove deep understanding of ML math and architecture, not just library usage.

Ann from Scratch for Spam Detection

A pure Python implementation of an Artificial Neural Network (ANN) for email spam detection, built entirely from scratch for educational and demonstration purposes. The project includes all steps: data preprocessing, text vectorization, normalization, dataset balancing, and custom implementation of forward and backward propagation. No machine learning or deep learning libraries (like scikit-learn, TensorFlow, or PyTorch) are used everything is coded manually to help understand the inner workings of neural networks. This project is featured on my portfolio to demonstrate how classic machine learning can be implemented and visualized without external libraries. If you want to experiment yourself, you can download the dataset here.

×

View Code