Mandavalli N V Saisandeep

Kaleidoscope (Retrieval-Augmented Chatbot)

• Constructed ranking system to optimize relevancy of retrieved chunks in RAG model by prioritizing documents based on last updated date, analyzing title significance using cosine similarity with query, and gauging description relevance through cosine similarity.
• Designed and implemented a prompt-based workflow to classify questions into summary-based, fact-based, FAQ, or support inquiries. Based on the classification, the system directed each question to the appropriate processing path, ensuring accurate handling of follow-ups and regenerated queries.
• Replaced the prompt-based FAQ and support handling with a vector search approach using Redis. Stored questions as vector embeddings, with metadata, and used cosine similarity to find the most relevant match. This improved accuracy, handled question variations better, and reduced response time from 13 seconds to 8 seconds.
• Built a Trending Questions feature to cluster semantically similar user queries and identify the top three most frequent questions weekly, improving insight into user interests.

AI Agents in Salesforce.

• Developed and deployed AI agents on Salesforce CRM using AgentForce capabilities, integrating flows, custom objects, permission sets, agent topics, and actions to streamline user interactions and enhance CRM intelligence.
• Built AI agents capable of summarizing standard and custom Salesforce objects (Accounts, Opportunities, Contacts, Activities) and answering user queries based on uploaded documents (PDFs, PPTs, Playbooks) stored in the Data Library.
• Created a Tableau Next Concierge Agent to interpret natural language queries and return visual insights using Tableau metrics, dashboards, and semantic models, significantly improving data accessibility.
• Engineered an automation agent to extract quote and quote line data from uploaded images and update corresponding Salesforce records, automating manual quote creation processes.
• Developed dynamic account summary generation tailored to user roles and permission sets (e.g., CRM, Health, Public Sector), ensuring secure and personalized data delivery.