◆ GOWTHAM.SAIROLE ARCHITECTLOC BLR · UTC+5:30AVAILABLE

ARCHITECT · PLATFORMS · ML · REAL-TIME ANALYTICS

GOWTHAM
SAI

I build the platforms other teams build on: distributed systems, ML & agent platforms, and real-time analytics.

Most recently Architect at Zepto, where I built the ads, data & analytics, GenAI and ML platforms. Before that, ShopUp, Gojek and Voonik.

Next Exploring a company around autonomous agents for engineering orgs. dev-agent is where the conviction comes from.

View work →Résumé ↓
↗ GitHub↗ LinkedIn✉ Email
Gowtham Sai
SUBJECT gwthm.inBASE BLR

01 Impact readout

Ad revenue
100×
Growth on an in-house ClickHouse ads platform
Migration
1000+
VMs & services moved to Kubernetes at Gojek
Cost
−90%
In-house Mixpanel replacement, shipped in 3 months
Onboarding
7×
Faster Kafka consumer onboarding via the Macros framework
Reliability
−40%
Incidents cut via K8s runtime tuning & Envoy breaking
Latency
−95%
Lower latency moving a hot path from Python to Go

02 Selected work

RUNNINGPRIMARY PROCESS · AI PLATFORM · ORG-WIDE

dev-agent: an always-on AI teammate for the whole org

An autonomous AI agent that started as one team's internal tool and got pulled in across the org. Any department can onboard its own persona: it plugs into the tools they already use, remembers context between tasks, and handles the routine work they would rather not touch, quietly saving hundreds of engineering hours.

2,000+PRs raised & merged
15+Teams onboarded
100+Codebases contributed to

Where it helps

  • Engineering: triage, reviews, on-call assists
  • HR: policy answers, onboarding nudges
  • Fin-ops: routine reconciliations

Benefits

  • Onboards a new team in a day
  • Improves the more each team uses it
  • People approve anything high-stakes

Trade-offs

  • High-stakes actions need review
  • Quality rides on strong evaluation
  • Each new team takes some tuning
EngineeringHRFin-opsOn-callOrg-wide
Read the story ↗
Analytics · Revenue
Ads & Brand Analytics
Zepto's ads platform on ClickHouse: real-time ROAS, spend accuracy, sub-second dashboards. 50% faster queries.
ClickHouseOLAPGo
Read: dedup at scale ↗
ML Platform
End-to-end ML Platform
Onboarding → features → routing → inference telemetry, with async inference. Cut model onboarding 90%.
ML InfraAsyncTelemetry
GenAI · Reliability
Aegis: DR & RCA portal
GenAI portal for disaster resilience and business continuity, at 60% accuracy on automated root-cause analysis.
GenAIRCA
Dev Productivity
Macros framework
Standardized retries, concurrency & DLQ for async workflows. Kafka consumer onboarding from a week to a day.
KafkaDLQGo
Infrastructure
1000+ VM → Kubernetes
Led Gojek's transport migration: cluster setup, topology, and workflows adopted as company standard.
KubernetesEnvoyGCP
Read: no more CPU ↗
Cost · Analytics
Mixpanel replacement
Migrated to an OSS + ClickHouse stack in ~3 months, covering key use cases at 90% less cost.
ClickHouseOSS

03 Writing & thoughts

AIJul 2026We Built an AI Teammate the Whole Company Uses. The Hard Part Wasn't the AI.↗SystemsJul 2026Deduplication at Scale with ClickHouse↗SystemsApr 2025Everything Was Fine Until Kubernetes Said No More CPU↗ThoughtsJun 2021Discussion vs Argument↗
All writing →

04 · Open channel

Got a system that's fighting you?

Building in autonomous agents, weighing a founding engineer or architect, or just want to trade notes on systems at scale? I'm easy to reach. Open to the right role, too.

Email me →LinkedInGitHubRésumé ↓