I’m Vishnu Sujeesh. I work on the parts of AI that remain after the novelty wears off: retrieval, evaluation, architecture, infrastructure, and the judgment required to make them useful.
My career has moved steadily closer to the decisions inside a system: first improving business operations, then building platforms, then machine-learning products, and now leading enterprise AI.
Enterprise AI strategy, architecture, and delivery
Machine learning and full-stack product engineering
Full-stack platforms and developer infrastructure
Marketing activation, integration, and analytics
The tools changed—from analytics and integration to platforms, machine learning, and AI—but the work has remained the same: understand the real constraint, make the system clearer, and leave it more dependable than I found it.
My work sits between technical depth and practical consequence. The model matters. So do the data, interfaces, incentives, failure modes, and people around it.
A multilingual retrieval system combining lexical search, embeddings, reranking, evaluation, and production observability.
A behavior-driven system designed around relevance, long-term usefulness, and measurable user outcomes.
A reusable foundation for tools, orchestration, memory, guardrails, evaluation, and human review.
Architectural and workload changes that improved scalability while removing unnecessary cloud spend.
National University of Singapore
A foundation in engineering, biomedical systems, and analytics shaped how I approach technology today: as an interconnected system of people, constraints, evidence, and consequences.
Without a clear evaluation, most iteration is only motion.
Models and vendors change. Good interfaces, contracts, and feedback loops endure.
Automation should reduce cognitive load without hiding important decisions.
A simpler system that can be understood and operated is often the more intelligent choice.
An elegant system should eventually improve cost, speed, quality, or human capability.
Turning research notes into a grounded publishing workflow.
How lexical search, embeddings, reranking, and evaluation work together.
Designing autonomy around constraints, tools, feedback, and human judgment.