Technologies & Methodologies
I’m experienced in machine learning, deep learning, computer vision, NLP, and generative AI, where I’ve honed my expertise through years of hands-on projects. I’m also well-versed in text-to-speech, speech-to-text, and quantum computing foundations, alongside techniques like supervised and unsupervised learning, transfer learning, reinforcement learning, federated learning, MLOps, A/B testing, hyperparameter tuning, and data pipeline optimization, which I’ve applied to streamline complex systems.
Frameworks & Tools
Skilled in TensorFlow, PyTorch, Keras, Hugging Face Transformers, OpenCV, spaCy, LangChain, and LlamaIndex, with experience in Blender, Maya, and NVIDIA Omniverse for 3D simulations and Stable Diffusion / ComfyUI for generative content creation, driving innovative AI solutions.
Models
Deep expertise in transformers (BERT, GPT, T5), CNNs (ResNet, VGG), RNNs/LSTMs, GANs, diffusion models, and LLMs (LLaMA, Mistral, Mixtral, Dolphin, Claude, ChatGpt, Grok).
Agentic AI & RAG Experience
Developed a scalable multi-agent framework built initially in LangChain, and later Microsoft's AutoGen library, in the early LLM era, when models like GPT lacked native search capabilities. Designed as a conversational research assistant, it handles queries on topics like tech trends or education through dynamic agent collaboration. The system includes specialized agents: a query agent to parse input and orchestrate tasks; a search agent to fetch data from Google, YouTube (via transcript extraction), and Reddit; a synthesis agent to consolidate and summarize results; and a response agent (powered by GPT variants) to generate outputs. Agents interact via AutoGen's message-passing protocol, enabling feedback loops for iterative refinement and adaptability. To boost accuracy and minimize hallucinations, I integrated Retrieval-Augmented Generation (RAG): retrieved snippets are embedded (initially via custom LoRA fine-tuning) and stored in a FAISS vector database. As AutoGen advanced, I adopted its native tools for seamless RAG pipelines, feeding context directly into the LLM. This project enhanced my expertise in agentic AI, demonstrating how multi-agent systems and RAG enable robust, scalable interactions beyond single-model limits, such as real-time Q&A sessions.
Additional AI Applications
I’ve built candidate-matching algorithms and applicant tracking systems that streamline hiring, alongside automation scripts that saved over 1,000 hours. My work spans content creation for music, movies, and games, as well as workflow automation, process improvement, and game automation. I’ve tackled a bit of everything in automation, always finding ways to make systems smarter and more efficient, often cutting process inefficiencies by up to 80%.
