- Role
- Senior Software Engineer, Toosi
- Based in
- Bengaluru, India
- Experience
- 7+ years in production software & AI
- Education
- MSc ML & AI · PG Diploma ML & AI · B.Tech CSE
I'm Sanjeet Kumar.
AI Software Engineer. Senior Software Engineer at Toosi.
I'm an engineer from India who became fascinated with one recurring problem: How do we make software understand what someone wants and actually help accomplish it?
That question has followed me for seven years, through recommendation systems, knowledge graphs, machine-learning infrastructure, conversational AI, sales agents and voice systems. Today it is the whole job: at Toosi I took an agentic outbound platform from idea to paying clients, and I own every layer of it.
I'm happiest when I can move across boundaries. One day that might mean reasoning about an agent architecture. The next might mean fixing a data pipeline, designing an API, reviewing a React workflow, evaluating model behaviour or deploying the service.
I care less about whether something is classified as “ML,” “backend,” “AI” or “product.” I care about whether the system works, and whether a real person gets what they came for.
The path so far
Full timeline- 012018 — 2021
Krowd
Started as the company's first intern and left as SDE-1. Crawling infrastructure pulling ~40K entities per city, a restaurant recommendation system, a knowledge graph that grew from 20K to 70K nodes, and a Kafka cluster.
- 022021 — 2023
Text Mercato Solutions Pvt Ltd (renamed Cataloging.ai, now Rubick.ai)
Joined as MLOps Engineer and left as Tech Lead. An Auth0 API gateway built from scratch, ML services on Kubernetes serving 1K+ concurrent requests, and a catalog-enrichment platform tied to client revenue.
- 032023 — present
Toosi
Senior Software Engineer. Took an AI-native outbound platform from idea to 2 paying clients, with live campaigns across the US, UK and Ireland, owning architecture, agents, data, product and AWS deployment.
What I'm good at
End-to-end AI product engineering
Problem framing, architecture, data ingestion, agent design, APIs, frontend workflows, evaluation, deployment and production iteration. One owner from the first diagram to the last deploy.
Production agent reliability
Grounding, confidence scoring, approval gates, deterministic business rules, ground-truth datasets, observability and failure handling, so an agent can be trusted with real clients.
Full-stack system design
React and TypeScript UX, NestJS and FastAPI service boundaries, database modelling, queues and workers, third-party integrations and scalable cloud deployment.
How I think about AI products
01
Intelligence must lead to action
A good AI product should connect understanding to meaningful actions, not simply generate text.
02
Autonomy requires observability
An agent you cannot inspect, evaluate or interrupt is a liability, not an intelligent system.
03
Data quality beats clever prompts
Reliable intelligence requires reliable context, retrieval and underlying information systems.
04
Production is part of the model
Latency, cost, failure modes, UI, approvals and user trust are part of AI system quality — not implementation details after the fact.
Recognition & community
- 2nd prize, Sandbox Developthon 2018Among 350+ teams
- 2nd prize, Amazon Alexathon 2018E-waste management application
- Podium, National Robotics Championship 2016400+ participants from 11 states
- Workshop facilitator, School of AI HubballiML, AI, NumPy, Pandas and data-analysis sessions for 80+ attendees
Outside the terminal
I teach when I can, I like to be in the room when a client first uses something I built, and I'm usually reading about how organisms learn, which is slowly turning into a side project.