Omer Faruk S.

Hi, I am Omer Faruk — a Computer Vision expert with a PhD in the field and over 15 years of hands-on experience bridging academic research and real-world AI solutions. My work spans from implementing cutting-edge academic papers to deploying production-ready AI systems. 🎓 Academic & Computer Vision Background My research foundation is deep and practical: ➡️ Developed a GAN-based facial image augmentation approach using latent space exploration via GAN inversion — outperforming existing methods in the literature as part of my PhD thesis. ➡️ Performed 3D facial expression recognition using point clouds and surface patches, achieving superior emotion recognition accuracy by combining both modalities. ➡️ Published multiple papers on skin lesion classification and segmentation in top-tier journals, achieving state-of-the-art performance on well-known benchmarks in the medical imaging literature. 🤖 Generative AI (Image & Video) ➡️ Stable Diffusion, Flux Dev / Schnell / Pro, DALL·E, Midjourney, ControlNet, ComfyUI, HunyuanVideo ➡️ Fine-tuning Flux models using Replicate and Fal ➡️ Deploying generative pipelines on cloud GPU providers like RunPod and Modal 🧠 Large Language Models & AI Agents ➡️ Integrating and deploying GPT-4, Claude 3 (Haiku, Sonnet, Opus), Llama, Mistral, Falcon, Gemma ➡️ Building intelligent agent pipelines with LangChain, CrewAI, AutoGen ➡️ Custom Model Context Protocol (MCP) server development 🗣️ Voice AI & Conversational Agents ➡️ Production-grade voice agents with VAPI, Bland AI, Pipecat AI, Air.ai ➡️ Real-time audio via WebRTC and WebSockets ➡️ Text-to-speech integration: ElevenLabs, Coqui, Bark ➡️ OpenAI Realtime Audio / Speech-to-Speech API 🔍 RAG & Vector Databases ➡️ End-to-end RAG pipelines for document Q&A, semantic search, and knowledge-grounded chatbots ➡️ Vector stores: Pinecone, Weaviate, FAISS, ChromaDB, PGVector 🖥️ Local LLM Deployment ➡️ Model quantization — GGUF, GPTQ, AWQ, INT4/INT8 for memory-efficient inference ➡️ GPU/CPU offloading strategies for running large models on constrained hardware ➡️ Serving via Ollama, LM Studio, llamacpp, vLLM, Text Generation Inference (TGI) ➡️ Model sharding, layer offloading, KV cache optimization ➡️ Context window management for production-grade private AI deployments 🛠️ Tech Stack ✔️ Computer Vision → OpenCV, PyTorch, TensorFlow, Keras, Scikit-learn, NumPy, SciPy, Pandas, Matplotlib ✔️ Generative AI → Stable Diffusion, Flux, ComfyUI, ControlNet, HunyuanVideo, DALL·E, Midjourney ✔️ LLMs & Agents → LangChain, CrewAI, AutoGen, MCP, GPT-4, Claude 3, Llama, Mistral, Gemma, Falcon ✔️ Local LLM → Ollama, LM Studio, llama.cpp, vLLM, TGI, GGUF, GPTQ, AWQ ✔️ RAG & Vector DBs → Pinecone, Weaviate, FAISS, ChromaDB, PGVector ✔️ Voice AI → VAPI, Bland AI, Pipecat AI, ElevenLabs, Coqui, Bark, WebRTC ✔️ Cloud & GPU → RunPod, Modal, Replicate, Fal, HuggingFace If you're looking for a specialist who combines deep academic expertise in Computer Vision with modern Generative AI and LLM capabilities, let's connect. I'm ready to take on challenging problems and deliver quality results.