DevOps, Platform Engineering & AI Enablement
Velocity is free now. Direction is the new scarcity.
I'm Vagmi Mudumbai, CTO of Urai AI Corp. Agents can write code faster than any team can review it. I build the platforms, pipelines and guardrails that keep that code shippable, and I help teams put agents into their products without losing control of cost or security.
Author
Rootless Containers from Scratch
A book and reference implementation built from Linux namespaces, cgroups and overlay filesystems in Bash.
CTO
Urai AI Corp
In-app agents that solve problems by writing code in a sandboxed runtime instead of chaining tool calls.
Platform architecture
Tunecore
Split a 10-year Rails monolith into services that handle $1bn+ in artist payouts.
What I do
DevOps & Platform Engineering
Infrastructure, CI/CD and golden paths that hold up when agents write a growing share of your code.
- Terraform, Kubernetes and CI/CD for teams with product-market fit and fragile infrastructure
- Guardrails for AI-generated code: review gates, test and lint policy in CI, sandboxed agent environments
- Cloud cost reduction and migrations
AI Enablement
Putting agents into your product and your engineering workflow, with cost and security under control.
- Code-mode agents, harnesses and sandboxed execution (Firecracker, V8 isolates)
- Private model routing, LLM gateways and agentic search
- Rolling out coding agents across a team: workflow, review practice, spend tracking
Featured essay
Wiring against the vibe
Agents made typing cheap, but typing was never the bottleneck. What Accelerate, Kill It With Fire and Wiring the Winning Organization say about running teams in the AI era.
Read the essay
Recent writing
What I'm working through
Urai - Unreasonable effectiveness of code-mode agents
Structured tool calling stalls on multi-step work. Given a sandboxed JavaScript runtime instead, Gemini 3.8 Flash scraped a website, read a knowledge base and built a print-ready slide deck in five minutes for fifty cents.
jevlite - Experience report in training the Decision Model
Building jevlite's training set from Super-NaturalInstructions and synthetic support and SIEM tickets, the teacher-labelling bug that cost 13 hours, and the QLoRA run on Gemma 4.
jevlite - A decision model based on LLMs
An open source take on TypeSafe's Jev. jevlite scores a fixed set of options from a single forward pass over the logits, so it returns real probabilities and cannot invent an answer.
Selected work
Problems, what I did, results
Tunecore
Platform Architecture
- Problem
- A 10-year-old Rails monolith run by six junior engineers. Over half the tests failed, coverage was under 30%, and there was no CI/CD.
- What I did
- Set up CI/CD and Terraform, moved workloads from bare EC2 to ECS, shipped the Rails 4 to 6 upgrade, and rewrote the distribution service in Go. When Believe acquired Tunecore, I built an EC2 plus Lambda FFmpeg pipeline to absorb the migration burst.
- Result
- 70% lower infra cost on distribution8M tracks transcoded in under a month$1bn+ in artist payouts
AICrete
Infrastructure & Platform Engineering
- Problem
- Compute-heavy workloads and SOC 2 requirements that Heroku could not serve at a sensible price.
- What I did
- Migrated production to AWS EC2 on Compute Savings Plans with dev/prod isolation and CI/CD to ECS. Built a Go CLI that gave developers a Heroku-like workflow, and a Go Windows service that tunnels the on-prem ERP to the cloud app through corporate firewalls.
- Result
- Hosting cost cut from $6k to $1.9k a monthSOC 2-ready environment separation
Urai code-mode agents
CTO & Platform Architecture
- Problem
- Structured tool calling stalls on multi-step work. Every step is another inference round-trip, and raw payloads fill the context window.
- What I did
- Built uraiJS, a serverless JavaScript runtime that starts execution contexts in 20ms, and a Firecracker VM that cold-boots in 2 seconds. Agents write and run code against published libraries and load code-mode skills when relevant.
- Result
- At least 50% lower agent cost than structured tool callingScrape, research and build a slide deck in 5 minutes for $0.50
Let's build something durable.
Engagements start at $5,000 a month. Tell me what you're building and where it hurts.
Schedule a 30-minute call