AI-Powered Shoe Design Generator
A generative AI assistant that transforms abstract user conversations into context-aware shoe designs. Fine-tuned LLaMA 3.2 3B to generate precise Stable Diffusion v2 prompts, bridging the gap between natural language and visual output. Designed as a scalable solution for the footwear industry — from automated concept design to personalized retail applications.
Satellite Image Translation — Pix2Pix GAN
Custom-trained U-Net generator paired with a PatchGAN discriminator for real-time satellite imagery translation into map-style renderings. Deployed via Streamlit for interactive inference. Achieved model stability and visual coherence in complex multi-feature satellite scenes, targeting GIS and urban planning workflows at scale.
StyleAI — Personalized Fashion Recommendation
A 5-stage AI pipeline: YOLOv8 segmentation splits user photos into face and body crops → fine-tuned VLMs (LLaVA 7B + Gemma 3) analyze them → 5 parallel MLP encoders fuse outputs into a 512D user state vector → 2-layer RGCN Fashion Knowledge Graph scores recommendations → Llama 3.1 8B drives multi-turn conversation over WebSocket streaming.
Sales Intelligence — Dual-Agent System
Production dual-agent pipeline built with LangGraph: a deterministic Sales Agent (extract → query → retrieve → synthesize) plus a context-aware RAG Chatbot with query rewriting. Hybrid retrieval over Qdrant Cloud with dense + sparse vectors and cross-encoder reranking. Full OpenTelemetry instrumentation for production observability.
Image Generation with GANs
Explored advanced GAN architectures for high-fidelity image synthesis. Implemented and compared multiple GAN variants, studying training dynamics, mode collapse prevention, and visual quality metrics. Applied to real-world generative design tasks.
AI Feedback Analysis System
Intelligent feedback processing pipeline leveraging LLMs and NLP techniques for sentiment analysis, topic extraction, and actionable insight generation from unstructured user feedback data. Built for scalable enterprise deployment.