I am an AI Engineering professional focused on building smart, data-driven agentic solutions. I work with machine learning, software development, and retrieval-augmented generation (RAG) pipelines to turn complex data into useful insights and practical applications.
Trained an EfficientNetB2 classifier on 7,000+ MRI scans to classify tumor categories, achieving 97%+ validation accuracy with Grad-CAM overlays.
Architected a real-time voice agent using LiveKit for low-latency streaming, Deepgram for ASR, and Cartesia for TTS voice synthesis.
Cooperative multi-agent workflow using LangGraph for hierarchical search planning, deep-dive information retrieval, and auto-reviewing reports.
End-to-end multi-document RAG pipeline using LangChain, ChromaDB vector store, and Gemini API supporting source citations and session history.
Fine-tuned Qwen 3.5 (4B) Vision on multimodal domain datasets using QLoRA via Unsloth, cutting training GPU VRAM footprints by 70%.
Traffic monitoring application developed as a Final Year Project (FYP). Uses YOLOv8 for detection and FastAPI for live telemetry analytics.
Architected and optimized backend services using Python and FastAPI, reducing service response latency by 20% and enhancing endpoint throughput. Collaborated with cross-functional engineering teams to design AI deployments, and built robust validation, automated error handling, and structured logging workflows to decrease system downtime and improve debug observability.
Conducted predictive modeling audits, analyzed customer churn parameters, and automated analytical reports using Python and visual dashboard frameworks.
Built classification models and descriptive statistics architectures mapping business KPIs. Optimized batch data parsing speeds across localized SQL servers.
Designed and optimized predictive ML/DL models using Python, PyTorch, and TensorFlow, improving accuracy by 15% across core business metrics. Automated daily data ingestion, processing, and ETL pipelines using Linux shell scripting and cron jobs, reducing manual operational overhead by 30%. Performed rigorous exploratory data analysis on multi-million row datasets, visualizing insights with Matplotlib and Seaborn to guide executive product decision-making.
A historical exploration of the foundational research papers in computer science, from Alan Turing's theoretical computing machine to the modern Transformer architecture that powers today's LLMs.
Read ArticleAnalyzing why prompting is no longer the bottleneck in advanced AI systems, and how engineering iterative, self-correcting feedback loops (agentic workflows) is the new frontier.
Read ArticleA deep dive into the architecture of LLM agents, demonstrating why foundation models require sandboxed execution environments, memory systems, and tools to interact with the world.
Read Article