Srinivas Dharavath

###Developed a Multi-Agent System for Automated Third-Party Risk Management (TPRM)### • Automated the end-to-end TPRM process, including Inherent Assessment, Residual Assessment, and Vendor Response Validation, ensuring seamless risk identification, control mapping, and compliance validation. • Integrated Crewai tools such as FileReadTool for data extraction, FirecrawlSearchTool and FirecrawlScrapeWeb-siteTool for web-based compliance validation, and RagTool for real-time structured and unstructured data analysis, enabling accurate decision-making. • Built a scalable framework with QA agents to ensure quality assurance and consistency across workflows. The system automates tailored risk mitigation strategies, reduces human error, and supports adherence to industry standards and regulatory requirements. ###Developed a Realtime Automated Voice Survey System using Twilio and OpenAI### • Implemented outbound voice calls via Twilio, including configuration of caller ID, fallback URLs, and call status tracking, enabling automated customer interaction. • Established a real-time WebSocket connection between Twilio and the OpenAI Realtime API to facilitate real-time media streaming for a dynamic and responsive chatbot experience. • Designed dynamic conversation flow with system prompts, passing survey structure to the AI and enabling dynamic questioning, response validation, and graceful handling of interruptions during customer interactions. • Fine-tuned OpenAI model parameters, including turn detection, audio formats, voice settings, transcription models, and language, optimizing the performance and responsiveness of the chatbot. ##Engineered an Enterprise-Grade, Multi-Agent RAG Chatbot System## • Architected a Multi-Agent RAG chatbot system supporting 10+ formats.Developed 3 distinct conversational AI models - Normal RAG for document retrieval, Deep-Think model for complex queries, & Web-Search model for real-time information

**LLM Fine-Tuning:** Generated a 9.75M token dataset from 15 Indian law textbooks and performed LoRA continual pretraining on LLaMA3 8B using A100 GPUs for over 20 hours. **Document Processing Chatbot:** Built a chatbot leveraging Claude 3.5 to analyze and process user-uploaded documents or folders, delivering high-accuracy insights with contextual understanding. **Scalable Storage:** Integrated Amazon S3 for secure and scalable storage of user-uploaded documents, enabling efficient real-time data processing. **Interactive Frontend:** Developed a user-friendly frontend using Streamlit, facilitating seamless document uploads and intuitive interactions with the chatbot. **Metadata Extraction:** Designed a tool to extract metadata from uploaded files of various formats using GPT-4o and Claude 3.5 Sonnet, deployed via FastAPI. **LLM Deployment:** Deployed Qwen 2.5 7B model on an EC2 instance and as a SageMaker endpoint for inference, optimizing performance and reliability. **High-Performance Infrastructure:** Utilized an ml.g5.24xlarge instance in SageMaker for faster inference responses, implementing auto-scaling to handle concurrent requests effectively. **Backend Optimization:** Ensured robust and efficient deployment pipelines for AI solutions, enhancing scalability and responsiveness for diverse user needs.

Working as a Backend Python Developer and AWS Solution Architect. Specializing in serverless architecture, customer engagement systems, and optimization strategies. Proven track record in AWS services, Django, and Terraform.