Building Convexa, a Knowledge-Based AI Agent
Built a conversational agent that answers customer questions from company knowledge.
Hariharan V
Senior Product Manager, Nagent
From their time as
Senior Product Manager
Nagent • May 2026
Overview
At Nagent, I built Convexa, a conversational AI agent that answers customer questions using documents and web pages as its knowledge base. I defined the product decisions, guardrails, and QA use cases, then documented the work in a public teardown.
The Story
Convexa was designed to answer customer questions from a company's existing knowledge, including documents and web pages. I worked across the product definition and hands-on delivery rather than limiting my role to requirements.
I defined how the agent should behave, the scope of its answers, and the guardrails around the experience. The work also included QA use cases and detailed decision documentation. I treated the interaction itself as a product surface, with attention to whether customers would feel comfortable and understood while speaking with the agent.
I documented the decision process, guardrails, and product details in a teardown linked from my resume and GitHub. This gives prospective teams a direct view into how I approach an AI product rather than asking them to rely on a role description.
The agent was completed as working proof of product capability. Broader customer or commercial outcomes are not yet available in the supplied record.
Ownership Snapshot
Broad role
Led product development for a conversational AI agent.
Goal
Answer customer questions from approved documents and web content.
Direct ownership
Owned product decisions, guardrails, use cases, QA, and the documented teardown.
Team execution
Worked within Nagent and presented related AI product concepts to company leadership.
Topics
- Conversational AI
- Knowledge Bases
- AI Guardrails
- Product Design
- Quality Assurance
Talk to Hariharan about this story
Send a message