Define better behavior
Turn product expertise, business requirements, important exceptions, and merchant judgment into behavior the Agent can follow.
ABOUT KETUPA
We’re building the optimization system behind better AI commerce agents.
AI agents are becoming part of how consumers shop and how businesses buy—helping customers understand products, compare options, choose solutions, and decide what happens next.
But launching an Agent is only the beginning.
The harder work is making it better: turning product expertise and business requirements into reliable Agent behavior, improving what does not work, and keeping what does work reliable as products and systems change.
KETUPA is building a faster, more systematic way to do that.
OUR POINT OF VIEW
Better models are not enough. Better Agent behavior has to be built and maintained.
Commerce teams already have product data, product experts, business rules, Agent platforms, and increasingly capable AI.
Yet an Agent can still recommend too early, miss an important exception, communicate the wrong level of confidence, or lose behavior that previously worked.
The challenge is not simply giving the Agent more information.
The challenge is making the right knowledge turn into the right Agent behavior—consistently.WHAT WE’RE BUILDING
A connected system for continuously improving commerce agents.
KETUPA brings together work that is often fragmented across Product, Ecommerce, AI, implementation, customer experience, and QA teams.Improve weak behavior
Identify where the live Agent needs improvement and turn accepted findings into real changes in the existing Agent stack.
Protect good behavior
Validate what works and keep important behavior reliable as products, models, prompts, catalogs, policies, and platforms evolve.
Define better behavior. Fix weak behavior. Protect good behavior.
WHY KETUPA
Better Agents. Faster improvement. Less repeated work.
KETUPA works directly with the Agent and commerce stack teams already use.We help define what should improve, evaluate how the live Agent performs, implement or co-implement the changes, validate the result, and continue maintaining important behavior over time.Better customer experiences
Help customers receive more appropriate guidance, recommendations, and next steps.
Faster optimization
Move from an unclear Agent problem to an implemented improvement faster.
More reliable change
Know whether the Agent became better—and whether behavior that already worked remained intact.
Less repeated work
Reuse accepted guidance, evaluation, and previous improvements instead of rebuilding the same work after every change.
KETUPA is hands-on today and increasingly software-driven over time. Every engagement helps make the next improvement faster, more reusable, and less dependent on repeated manual work.
LEADERSHIP
AI commerce behavior expertise, backed by production AI leadership.
KETUPA combines deep research into how AI Agents guide customers with senior experience building large-scale search, conversational AI, agentic systems, and production software.Our broader team adds specialist depth across evaluation, implementation, product quality, UX, customer success, QA, and software engineering.
Bufan Shen
Founder & CEO
AI Commerce Agent Research · Agent Decision Behavior · Product Direction
Bufan leads KETUPA’s vision, research, and product direction.
His work focuses on how commerce Agents interpret customer needs, apply merchant product knowledge and business rules, make recommendations, handle uncertainty and exceptions, and remain reliable as products and customer context change.
At KETUPA, he turns that research into systems that help merchants define, improve, and maintain better Agent behavior across B2C and B2B commerce.
Jian-Tao Sun, PhD
Co-Founder & CTO
Former Microsoft Partner Engineering Manager · Principal AI Engineering & Science Leader
Jian-Tao leads KETUPA’s technology strategy, architecture, engineering, and productization.
Jian-Tao is a senior AI engineering and research leader with more than a decade at Microsoft across Azure AI Search, enterprise conversational AI, Bing Search, retrieval, RAG, knowledge systems, and agentic search.
At KETUPA, he leads the technical platform that turns commerce behavior standards, evaluation, implementation, and reliability workflows into scalable production systems.
Varshith Dupati
Founding Member · AI & Software Engineering
Amazon Production Systems · Applied AI · Distributed Systems
Varshith builds the software and production systems behind KETUPA.
His work spans distributed systems, agentic workflows, RAG, evaluation, automation, MCP-based agents, continuous monitoring, and production reliability, with experience across Amazon-scale backend systems.
At KETUPA, he develops the platform, integrations, and automation that turn the company’s commerce-Agent methodology into scalable production workflows.
A senior cross-functional team built for the full Agent improvement lifecycle.
Beyond the founding leaders shown above, KETUPA includes senior specialists across AI evaluation and reliability, Agent implementation, product quality and UX research, customer success, QA and regression, and software engineering.The team works as one delivery unit—from understanding merchant needs and designing better Agent behavior to changing the live system, validating the result, and maintaining reliability over time.AI COMMERCE AGENT OPTIMIZATION & RELIABILITY
Make better Agent behavior easier to create, improve, and maintain.
KETUPA is building the technology, expertise, and operating system needed to continuously improve AI commerce agents as they become a larger part of how consumers shop and businesses buy.