infrastructure · devops · ai engineering consulting
Your infra is on fire.
I put it out, then stop it from reigniting.
Broken Kubernetes, gateways that won’t route, runaway cloud bills, AI stuck one step short of production. That’s the whole job.
- ▹Toptal Top 3% talent
- ▹Infra engineer @ TrueFoundry
- ▹10+ years shipping & breaking production
- ▹Author of the Lazy SRE security series
the stack I live in
services
What I help with
Infrastructure Debugging & Incident Response
Production is down, a pod won't start, or nobody knows why latency tripled. I debug it to root cause and get you back up.
learn more →API Gateways & Networking
Ingress that won't route, an API gateway nobody understands, or a VPN appliance you want gone. I make traffic flow the way it should.
learn more →DevOps & Platform Engineering
Kubernetes set up properly, infrastructure in code, and CI/CD that deploys without drama. The platform your team wishes they already had.
learn more →Cloud Cost Optimization (FinOps)
Your AWS or GCP bill keeps climbing and nobody knows exactly why. I find the waste and cut it, and your system stays just as fast and reliable as before.
learn more →Security & SRE Hardening (DevSecOps)
Shrink your attack surface and stop the 3am pages. Security and reliability hardening that respects your team's time.
learn more →AI Engineering & AI Agency
You want to ship an AI feature, a RAG assistant, an agent, a self-hosted model, but the gap between a demo and production is wide. I close it.
learn more →MLOps & Model Deployment
Your model works in a notebook. Getting it served, scaled, monitored, and safe to update in production is a different job. That is the one I do.
learn more →RAG Systems & AI Chatbots
You want a chatbot over your docs that gives correct answers with citations, but the prototype makes things up and nobody trusts it. I fix that.
learn more →AI Agents & Workflow Automation
Your agent works 70% of the time, which is a great demo and a terrible product. I get it to a reliability number you'd actually put in front of users.
learn more →LLM & AI Cost Optimization
Your OpenAI or Anthropic bill is climbing and a lot of it is waste. I find where the money goes and cut it without wrecking quality.
learn more →selected work
Problems I’ve already solved
Cut a SaaS startup's AWS bill by ~42%
A Series-A SaaS team's AWS bill had tripled in a year and nobody could say why. A focused FinOps pass cut it ~42% without touching reliability.
Cloud Cost Optimization Prod restored in under 1 hour, root cause in writingRecovered a production cluster from a CrashLoopBackOff outage
A node upgrade left an entire production namespace in CrashLoopBackOff. Mitigated in under an hour, root-caused to a probe and config-map mismatch, and fixed so it can't recur.
Infrastructure Debugging & Incident Response ~70% lower inference cost vs. all-API baselineShipped a self-hosted RAG assistant with an LLM gateway
A team with a promising RAG demo couldn't ship it. Accuracy was unmeasured and API costs were unpredictable. A measured pipeline behind an LLM gateway with hybrid serving made it production-ready and ~70% cheaper.
AI Engineering & AI Agency · MLOps & Model Deploymentinsights
Field notes from production
- ▹ EKS vs GKE vs Self-Managed Kubernetes: Where the Cost Actually Goes
- ▹ Ingress vs Gateway API: Planning the Move Now That ingress-nginx Is Retired
- ▹ Choosing a Vector Database for Self-Hosted RAG: What Actually Decides It
- ▹ A Zero-Downtime Kubernetes Migration Runbook That Survives Contact With Production
- ▹ Karpenter vs Cluster Autoscaler: Which One Should Be Scaling Your Nodes?
- ▹ Kubernetes OOMKilled: Debugging Exit Code 137 Without Just Raising the Limit
- ▹ Your AI Agent Works 70% of the Time. Here Is What the Other 30% Is Made Of.
- ▹ How to Cut an LLM API Bill 50-70% Without Anyone Noticing a Quality Drop
- ▹ Debugging Kubernetes Ingress: What 502, 503, and 504 Are Actually Telling You
- ▹ Self-Hosted RAG in Production: The Issues That Never Show Up in the Demo
- ▹ Kubernetes Cost Optimization: Where the Money Actually Goes and How to Cut It
- ▹ Kubernetes CrashLoopBackOff in Production: A Debugging Playbook That Actually Works
what clients say
Results, in their words
Quotes are anonymized. Most of this work happens under NDA.
Our AWS bill dropped ~42% in three weeks, and nothing got slower. He found the waste we’d stopped seeing.
He pulled us out of a CrashLoopBackOff outage at 2am, then left us with a runbook so it never happened the same way twice.
Our AI support agent deflects ~60% of tickets now, and it finally stopped making up answers.
Migrated us off a brittle VPN to a clean mesh and re-did the gateways. Zero downtime, no drama.
how it works
How an engagement runs
Reach out
Tell me what’s broken, or what you’re trying to build. A short message or a 30-minute call works.
Diagnose
I find the root cause, or scope the build, and write up a plan in plain language.
Fix & hand off
I do the work, narrate as I go, and leave your team with docs, runbooks, and clear ownership.
Questions before you reach out
What do you actually do?+
I’m an independent infrastructure, DevOps, and AI engineering consultant. I debug broken Kubernetes and cloud setups, design gateways and networking, cut cloud bills, harden security, and ship production RAG/LLM systems.
Do you do one-off fixes or longer engagements?+
Both. Emergency incident calls, short scoped projects like a cost-cutting pass or a gateway setup, and longer contract, fractional, or staff-augmentation work where I embed with your team. We pick whatever fits the problem.
How do I start?+
Book a free 30-minute call, or send a message with your stack, the symptoms, and your timeline. You’ll leave the first call knowing what’s wrong and how I’d fix it.
Who am I working with?+
Harshit Luthra, a Toptal Top 3% infrastructure engineer currently at TrueFoundry, with 10+ years of breaking and fixing production. You work with me directly.
Got a problem worth a look?
Book a free 30-minute call. We diagnose it together, and you walk away with a plan you can act on. You’ll get a straight read either way.