About

I lead the 12-engineer team that builds Generate, the enterprise AI platform at Iterate.ai, and I moved the team to an AI-native SDLC. I architected Generate from scratch, and it won AI Product of the Year from both Pinnacle and TMC.

Before managing, I built production AI/ML systems, from multi-agent platforms to edge inference.

Engineering Leadership · AI-Native SDLC · RAG · Multi-Agent Systems · Edge & On-Prem Inference · Enterprise AI Security

Impact at a glance

6→12
Team doubled
1 week
Per task, was
4 to 5 weeks
65%+
More merged PRs,
up to 95%
33%
Lower infrastructure cost

How I lead

Lead the move to an AI-native SDLCI moved the team from Windsurf to Claude Code in VS Code and rebuilt our lifecycle around it. Agents do the building; engineers stay accountable for every judgement call.
Review the spec, not the diffEvery change starts as a written spec, and the agent's plan is approved before any code is generated. Judgement goes in while a change is still cheap.
One engineer, end to endThe spec lives in the issue, the merged PR closes it, and the release tags itself, so each stage hands the next a versioned record. In one monorepo, an engineer and their agents own a feature from spec to release.
Stay close to the people with the problemI write requirements with customers and partner engineers at IBM, NetApp, and Intel, so little is lost in translation.
Grow leaders from insideSenior roles go to people who grew into them here. Tech leads own feedback and performance on their teams.

Experience Full history on LinkedIn →

Software Engineering Manager

Sep 2024 to Present

Iterate.ai, Toronto, ON

  • Doubled the team from 6 to 12 in a year, sourcing and closing 6 hires. I manage it across frontend, backend, DevOps, and AI/ML through 2 tech leads and 2 project managers.
  • Leading Generate's storage editions with NetApp, our largest current direction: permission-aware retrieval over customer data on NetApp ONTAP, so answers respect each user's file permissions without moving the data.
  • Set the roadmap that made Generate Iterate.ai's primary enterprise revenue product, and led the engineering behind two enterprise contracts and the healthcare launch with IBM, featured in IBM's product blog.
  • Promoted 2 engineers to Senior Applied AI Engineer and mentored 7 more. One former intern now owns the agents platform.
  • Took the team AI-native with Claude Code, spec-driven development, and one monorepo. A task that took 4 to 5 weeks now ships in a week, and merged PRs are up 65% to 95%.
  • Measured the gain by task type: about 10× on scaffolding such as auth, UI components, and API schemas; 2× to 4× on complex logic and bug fixes in the large codebase; and up to 30× over our engineers on sandboxed document-handling optimization.

Data Scientist

Nov 2023 to Sep 2024

Iterate.ai, Remote, Canada

Earlier ML Engineering

2021 to 2023

ExentAI & Simplyfai, Sri Lanka (Remote)

  • Multilingual OCR over 65,000+ historical newspapers, and machine fault prediction that cut unplanned downtime 67%.

Case study: Generate Iterate.ai →

I have owned Generate's architecture and infrastructure budget from a 6-engineer edge project to a multi-tenant SaaS product.

  • ProblemRegulated customers cannot send data out. Air-gapped installs had to keep every feature.
  • DecisionOne system everywhere. The same microservices run in cloud, on-prem, and on edge hardware down to a laptop.
  • DecisionIsolate every agent session. Sessions run in kernel-level sandboxes, and retrieval respects each user's file permissions.
  • DecisionGate every model. A 12-point compatibility suite runs before any model goes live.
  • Outcome99.8% uptime at 50,000+ requests/second. Consolidating services cut infrastructure cost 33% and made the platform 4× faster.
  • Outcome94% document extraction accuracy on nested tables and mixed formats.

Writing & talks All posts →

Research & Open Source 260 citations →

Google Scholar: h-index 7 · i10-index 6 · 4 US patent applications

Selected publications
Open source
  • MCPStepgate: keeps an agent on one step until checks pass. 13 stars, 2 forks, 1,270 downloads.
  • ModelMedBERT: 570,001 downloads on Hugging Face.
  • PyPINERP: fine-tune transformer NER models. 72,209 downloads.

All publications, patents, and BibTeX →

Recognition & Education

  • 2026CRN AI 100: Iterate.ai among the Top 20 Hottest AI Software Companies.
  • 2025Pinnacle and TMC AI Product of the Year, both awarded to Generate.
  • 2022Top 5 Finalist, HPCIC Innovation Challenge with team Medispeech@NTU.
  • 2018National merit scholarships: Suba Pathum, Mahapola, Ceylinco Pranama & Dialog.
Education
  • 2025MASc, Electrical & Computer Engineering, McMaster University. Graduate Research Scholarship; thesis on retrieval for scientific documents.
  • 2023BSc Eng (Hons.), Computer Science & Engineering, University of Moratuwa. First Class Honours.
© 2026 Charangan Vasantharajan