- Architect and implement Large Language Model (LLM)-powered systems for automation, conversational interfaces, and data-driven workflows.

- Design and deploy Retrieval-Augmented Generation (RAG) pipelines using FAISS/ Qdrant for high-recall semantic search.

- Build AI agents capable of structured reasoning, SQL interaction, and multi-step task execution.

- Develop FastAPI-based inference services for multi-user environments with asynchronous request handling.

- Design modular LLM abstraction layers enabling model interchangeability (API-based and local models).

- Deploy GPU-accelerated local LLMs for secure and low-latency environments.

Stack: Python, FastAPI, Aiogram, Qdrant, FAISS, AsyncIO, PyTorch, HuggingFace, Transformers, LoRA, QLoRA, Sentence Transformer

- Architected modality-specific deep learning models for image and audio deepfake detection.

- Defined training pipelines and evaluation protocols using confusion matrix-driven optimization.

Stack: Python, Tensorflow, Keras, Numpy, Colab

Covered by NDA

Focused on early adoption and applied integration of Large Language Models into automation workflows.

- Developed Telegram-based AI assistants powered by LLM APIs.

- Built document summarization and text-processing automation tools.

- Designed prompt-driven pipelines for structured output generation.

- Implemented backend services for LLM-based processing using Python.

Stack: Python, OpenAI, FastAPI, SQLite

- Designed and trained 3D Convolutional Neural Network (3D CNN) architecture from scratch for event detection from video sequences.

- Implemented real-time inference pipeline (~30 FPS).

- Built multi-threaded evaluation system for accelerated validation.

Stack: Python, Tensorflow, Keras, Numpy, Multithreading

- Designed and implemented CNN.

- Defined model topology, convolutional blocks, and training strategy.

- Built preprocessing and augmentation pipelines in Python.

Stack: Python, Tensorflow, Keras, Numpy, Multithreading