I work across the full AI delivery stack: data pipelines, experimentation, model fine-tuning,
inference optimization,
deployment, and monitoring. My recent work spans generative AI systems, enterprise chatbots,
speech-to-text pipelines,
semantic search, video intelligence, and production recommendation engines.
The focus is consistent: ship systems that are reliable, cost-aware, and tied to business
outcomes rather than demos.
That includes LoRA/QLoRA fine-tuning of open-weight LLMs, RAG pipelines with FAISS and Pinecone,
real-time inference,
and ML platform practices including CI/CD, MLflow, and experiment tracking.
Deep learning work is built on PyTorch — from transformer architectures to distributed training
across multi-GPU setups.
Applied to fine-tuning LLaMA and Mistral models, building vision encoders, and optimizing models
for production inference
with quantization and dynamic batching.
On the infrastructure side: Docker, Kubernetes, AWS, and GCP Vertex AI for ML workload
orchestration, with automated
monitoring and retraining pipelines. Machine learning system design — from data versioning
through model registry to
serving — is as much the job as the models themselves.