2025
AI/MLAI Customer Support Agent
Intelligent CS agent with text and voice integration that can answer customer queries. Web-scraped company data into a Pinecone vector database and semantically searched context via RAG.
Overview
A full-stack web application that answers questions about Aven via text and voice. The system performs web scraping of company materials, embeds the cleaned text, and stores vectors in Pinecone to enable Retrieval-Augmented Generation (RAG). It includes an evaluation suite of ~50 user questions, safety guardrails for sensitive/unsafe content, and meeting-scheduling logic that parses natural-language date/time.
Challenge
Building a production-ready AI customer support system with voice capabilities and safety controls required addressing dispersed and inconsistent public information, ensuring semantic relevance, real-time voice integration, implementing proper guardrails, and accurate scheduling across locales and time zones.
Solution
Developed a comprehensive customer support solution with a modular RAG pipeline (scrape → clean → chunk → embed → upsert to Pinecone → retrieve → answer with citations), multi-modal interface supporting both text and voice using OpenAI VAPI, comprehensive safety guardrails combining pattern checks and LLM moderation, and smart scheduling that detects intent and extracts date/time.
Impact
Delivered a production-ready demo with reliable text and voice chat in two weeks. Achieved significant gains in answer relevance through chunking and top-k tuning, established repeatable evaluation and safety controls for ongoing improvement, and prepared a clear path to integrate real calendar scheduling and richer analytics.
Technologies
YEAR
2025
CATEGORY
AI/ML
