Lavanya Govindaraju

Developed 30+ psychometrically validated assessment items for professional LLM certification, covering advanced topics including parameter-efficient fine-tuning methodologies (LoRA, QLoRA), RLHF, and DPO. Conducted technical validation and peer review processes to ensure assessment alignment with production LLM workflows, encompassing deployment constraints, evaluation frameworks, benchmarking methodologies, and safety considerations.

Architected AI system for women's health prediction combining LSTM-based time-series forecasting on synthetic dataset across 6 biomarkers with LLM-powered conversational interface and semantic RAG pipeline. Built end-to-end MVP with FastAPI backend, TypeScript frontend, and Weaviate vector database, implementing structured prompt engineering, multi-step agent workflows with LangChain, and Langfuse for continuous evaluation. Developed agent orchestration system coordinating prediction generation, knowledge retrieval, safety validation, and response synthesis, establishing modular architecture designed for future microservices scaling and regulatory compliance.

Developed intelligent photo system with chat-based UI integrating vision-language models and LLM-based refinement through RAG architecture, achieving 43% caption quality improvement validated through user study across 1,000 images with 10 participants. Designed hybrid retrieval pipeline using LangChain, combining FAISS-indexed semantic search with GPT-4 and Llama contextual generation, implementing production Flask REST APIs with JavaScript frontend, Docker containerization, and CI/CD automation.