Sreeraj Marar

I build production AI agents and the platforms they run on.

Now I’m building an AI agent platform at a stealth startup in New York. Before that I was an agent engineer at Sierra, shipping voice and chat agents for financial institutions, and a forward-deployed engineering lead at Gecko Robotics, building its agent platform for power plants.

See the work
Remaining wall --.-- mm Nominal 12.70 mm
A wall-thickness map from a robotic ultrasonic inspection. Blue-green is sound steel; orange has worn below its minimum allowable thickness. Move across it to take a reading.
A refinery lit up at night through falling snow, across dark water.
Refineries and power plants run through the night. So does the software that watches them.

About

Sreeraj Marar

I studied computer science at the University of Michigan, with a minor in entrepreneurship, and started my career as a software engineer at Armorblox (acquired by Cisco), building threat detection and reporting infrastructure that helped protect more than a million enterprise mailboxes.

At Gecko Robotics I was a forward-deployed engineering lead. I managed engineers and served as a technical lead within Gecko’s $100M partnership with NAES, working with C-suite executives to modernize US power plants with AI and robotics. I led Cantilever AI, an agentic platform that uses RAG over OEM manuals and plant context to triage power-plant anomalies for inspection, maintenance, and asset-health work, and owned Asset Manager, which helped close Gecko’s first SaaS deal. I also shipped the inspection-data pipelines, on GCP with Spark and BigQuery, that processed more than a hundred million readings from robots crawling boiler walls.

At Sierra I was an agent engineer, owning voice and chat agent delivery for four financial-services customers at once, including Fiserv and FINRA. I built and launched a production voice agent for dispute management in 14 days as the sole engineer, and designed the multi-org authentication and CI/CD that let a customer team of about 20 engineers co-build agents in a regulated environment. I iterated every agent with simulated conversations, LLM-as-a-judge evaluations, and production call review; containment typically settled around 80%.

Now I’m building an AI agent platform at a stealth startup.

Work

Pick a role to change what the monitor shows.

An operator at a wall of annunciators, strip-chart recorders and meters in a 1967 power station control room.
Callide Power Station, about 1967. Someone still has to decide what the readings mean.

Sierra

2026

Agent Engineer

Voice & Chat Agents

Production AI agents for customer experience at financial institutions including Fiserv and FINRA, handling voice and chat conversations with real end users.

Gecko Robotics

2022–2026

Forward Deployed Software Engineer, Lead

Cantilever AI

Enterprise agent deployment platform for industrial operations. RAG, multi-agent orchestration, and custom fine-tuned models that help power-plant operators make proactive decisions, shipped through a strategic industry partnership.

Asset Manager

Industrial asset health & capital management platform on Cantilever, Gecko’s first commercial software offering. Interactive 3D inspection visualizations rendered across more than a hundred million data points.

R-AUT Reporting Engine

Automated pipeline transforming high-resolution robotic ultrasonic inspection data, 100–1000x denser than legacy scans, into validated digital and PDF reports, reducing prep time ~90%.

Armorblox (acquired by Cisco)

2021–2022

Software Engineer

Email Threat Detection

Threat detection and reporting infrastructure at an AI email-security startup, acquired by Cisco, with real-time analysis protecting over a million enterprise mailboxes.

Contact

Tell me what you’re building.

New York City

Starting

Drag to look 1–4 sections Esc back