At VISTA Lab, I designed and deployed an end-to-end AI generation framework, including a RAG pipeline over 400,000 clinical trial records. The semantic search system achieved 95%+ recall, with a 6,000x speedup over exact search.
I also fine-tuned Qwen2-7B-Instruct with LoRA on an NVIDIA A100 via SLURM, reaching a BERTScore F1 of 0.8358. I built the MLOps workflow with DVC, TensorFlow, Weights & Biases, and MLflow, and instrumented LLM pipelines with Langfuse and LangSmith.
Previously, I built and maintained Python/Django applications at Realmind Technologies and deployed containerized services on GCP/GKE and AWS EC2. At Virtual Intelligence Solutions, I migrated a CI/CD pipeline to GitHub Actions and optimized Techcabal’s database queries and caching strategy, improving page load speed by 60%.

