Portfolio

Work that shipped.

Real projects, real production systems, real numbers. Every metric below came from an engagement I designed, built or optimized end-to-end.

DevOps

Manual deployments took 2 engineers 4+ hours per release. Bi-weekly releases out of pure risk aversion.

4 hrs → 8 min deployBi-weekly → 50+/week1 hr → 30 s rollback

GitHub ActionsArgoCDEKSTerraformHelm

AI Engineering

50,000+ medical documents needed to be searchable for clinicians. Keyword search was useless and off-the-shelf RAG blew the latency budget.

92% retrieval accuracy<800ms p50 latency5,000+ queries/week

LangChainPineconeOpenAIpgvectorFastAPI

Cloud

30+ microservices on fragile EC2 — weekly incidents, manual scaling, risky deploys. Zero-downtime migration required.

Zero-downtime migrationWeekly → 2 incidents/qtr25% infra cost cutSOC 2 Type II: 0 findings

AWS EKSIstioTerraformKafkaPostgreSQL

Cloud

$82K monthly cloud bill growing 15% MoM. No tagging, no cost allocation, no guardrails — nobody knew where the money went.

$82K → $49K (40%)127 orphaned resources killedFull ROI in <1 month

AWSTerraformLambdaPythonCloudWatch

AI / MLOps

The ML team spent 60% of their time on infra plumbing. Setting up a single training run took 3 days.

3 days → 30 min setup2× training throughputGPU util 30% → 75%

MLflowGKEAirflowS3Docker

Cloud

No DR plan at all — a single-region outage meant total downtime. Investors required DR capability for the Series B round.

RTO < 1 hourRPO < 5 minutes0 data loss in 2 drillsCited in Series B close

AWSRoute53RDS Cross-RegionTerraform

Platform

Every team had its own deploy process. New services took 2+ weeks to reach production and onboarding a new engineer took 3 days.

2 weeks → 2 hrs to prodOnboarding 3 days → 1 day45 services in month one~40% less platform toil

BackstageKubernetesArgoCDTerraformGitHub

Want results like these?

Book a free 30-minute call to talk through your project. I’ll tell you honestly whether I can help.