Sai Vignan Malyala

Designed and deployed a multi-agent Generative AI system to enhance clinical trial inspection processes by aggregating structured and unstructured data sources for areas like protocol deviations and adverse events. Integrated parallel generation, semantic caching, agent memory, and human-in-the-loop mechanisms to improve system performance. Conducted regular feedback sessions with client stakeholders to gather insights on usability and emerging use cases. Categorized feedback into key areas (usability, performance, training needs) for structured analysis. Implemented a continuous improvement loop, refining AI models and processes based on categorized feedback, issue prioritization, and regular progress updates. Delivered structured reports on feedback trends, system adjustments, and key recommendations for performance optimization. Coordinated cross-functional collaboration across stakeholders and technical teams to ensure tool refinements aligned with business goals.

Built AI and Gen AI products effective for massive real time usage. Used RAG, LLMs, AI agents, langchain, langgraph, custom retrievers, fine-tuning of smaller LLMs, evaluation of results, Azure for deployment. Engaged stakeholders through feedback sessions, encouraging participation in feedback mechanisms like in-app feedback surveys and regular check-ins. Compiled a comprehensive use case inventory, associating user feedback with potential improvements and scaling opportunities. Led feedback analysis sessions to align stakeholders on system challenges and implemented an iterative response loop for refinements.

Led a team of 4 AI developers. Fine-tuned multiple LLMs on scale with multi-GPUs (minimum 2 nodes, each node has 8 GPUs, each GPU is A100 40GB RAM). Guided for Gen AI applications such as Custom LLM model finetuning, RAG generation, Extractive QA search, and Semantic Search applications. Facilitated regular progress reviews, presenting structured insights and improvement suggestions based on qualitative feedback analysis from internal users.

Identified First party audience on the platform based on different health predictions and classified users based on Social determinants for web, android & IOS using app usage & behavioural data. Worked on Large scale data modelling of around 400 features and 600M records. Worked with Pyspark, EMR, Sagemaker pipelines, python, Aws Athena, gin configs, etc. Predicted users on Home owners, breast cancer, business travelers, Psoriasis, Asthma based on app usage data (finding soft labels based on data distribution). Led the team of 3 working on SDoh use-cases. Applied Generative AI on Weather article checks with multiple data points. Built a Vector search platform for stakeholders that ingests data, finds right videos and articles based on search inputs and maps it to right tags.

Implemented Table Detection and Table Structure Detection on banking documents using CascadeTabNet, table transformer and LayoutLM. Developed cost-saving solutions for native PDF extraction and OCR models. Achieved 94% accuracy comparable to premium API solutions.

Led development for Resume and JD parsers achieving 84% F1-score processes. Implemented custom scraping engines, chatbots, and data analytics use cases. Built actionable NLP-based recruitment tools, reducing operational API costs. Delivered scalable products capable of handling high throughput with low response latency.

Developed domain-based solutions for article generation in Sports & Betting. Built products from scratch including an LLM chatbot for recruiters and an advanced candidate validation tool utilizing Generative AI. Delivered article generation capabilities for media companies using LLMs. Conducted use-case analysis and domain scenarios before project implementation.