LLM Chatbot & AI Agent Development
I build production-ready AI chatbots and agents powered by GPT-4, Mistral, Llama, or Phi-3 — with RAG pipelines, memory persistence, and Text2SQL for querying your real data. Delivered with FastAPI backends, Docker packaging, and clean documentation. Ideal for enterprise knowledge bases, academic portals, or customer-facing support tools.

Voice AI & STT/TTS Pipeline
I design multilingual voice assistants with Whisper/Wav2Vec for speech-to-text and Bark/Tortoise for natural speech output — integrated with an LLM conversation engine. Built for real-time interaction, deployed on Azure or AWS, and orchestrated via Apache Airflow for automated model updates and monitoring.

Custom LLM Fine-tuning
Need a model that understands your domain? I fine-tune BERT, Mistral 7B, Llama 2, and other open-source LLMs on your data — achieving up to 30% accuracy gains on domain-specific tasks. Includes evaluation metrics, model versioning, and deployment-ready output via HuggingFace or custom APIs.

MLOps & Cloud Deployment (AWS / Azure)
I set up end-to-end MLOps infrastructure: CI/CD pipelines via GitHub Actions, Dockerized model serving, Kubernetes orchestration, and model monitoring on AWS or Azure. I can also evaluate AWS Bedrock models (Claude, Titan, Llama) for your use case and build the deployment pipeline around the best fit.

Computer Vision Solutions
From real-time object detection with YOLOv5 on edge devices (Raspberry Pi) to medical image classification, I build CV systems optimised for your hardware and accuracy targets. Past results include 85%+ real-time detection accuracy and 95% on benchmark classification datasets.

Data Science & Predictive Analytics
I build predictive models for finance, healthcare, and energy — from loan default detection (XGBoost/Random Forest) to renewable energy forecasting (92% accuracy). Includes full pipeline: data cleaning, feature engineering, model training, evaluation, and a FastAPI or Flask deployment.