One human.
Eleven agents.
Krish Mehta runs a one-person studio where AI agents do the engineering. He writes the brief, sets the bar, and decides what stays.
TripOS.
The AI concierge that takes over once your trip starts. I'm the only human in the loop — eleven agents ship the product.
“Built solo over weekends. Real card-counter depth — Hi-Lo, KO, Illustrious 18, Wonging.”
Blackjack
Trainer.
A complete product, shipped solo, with the depth a serious player needs.
I think like a founder. The taste, the brief, the call on what ships — that's the work I do. The agents handle the code.
By day I'm at Harness, leading enterprise implementations. By night I run a one-person studio. TripOS proves an operator with strong taste and the right tools can ship a real product without an engineering team.
PayPal taught me to keep score. I surfaced $5M in audit savings with an ML model, saved $60K/month on Artifactory storage, and cut Artifactory downtime 40% by moving it to Kubernetes.
- Harness2025 —
- PayPal · SWE2024 — 25
- PayPal · Intern2023
I build agents that ship to production.
Eleven named agents ship TripOS, coordinated by a control plane, each pinned to a model tier — Opus, Sonnet, Haiku — chosen on cost and workload. I work on the application side of AI: MCP servers, agent orchestration, LLM-augmented tooling. Not training; integration.
- claudeopus-4.7
- claudesonnet-4.6
- claudehaiku-4.5
- openaigpt-runtime
- mcpcustom-nlp
11-agent control plane. Cost-aware model tiering across Opus, Sonnet, Haiku. Production traffic at tripos.dev.
Pure-function game engine with invariant tests. Spaced-repetition learning loop. Built with AI but product is deterministic.
Custom MCP server unifying search and command across Artifactory, GitHub, and Jira — material SDLC velocity gain.
Enforces enterprise governance and coding standards across PRs. 100% compliance with internal rules.
Jira task analysis model surfaced $5M+ in cost savings for the Finance audit team.
Three accelerators with Claude + GitHub + Harness — incl. a one-shot repo → CI/CD framework with STO.
I build agentic systems for real production. TripOS has 11 specialized agents (ATLAS, FORGE, PILOT, CONCIERGE…) coordinated by a control plane, each pinned to a specific model tier and budget.
At PayPal: custom NLP MCP server bridging Artifactory + GitHub + Jira, an ML model predicting Jira task costs (surfaced $5M+ in audit savings), a GenAI PR-review bot enforcing internal coding standards across 20,000 developers.
At Harness I'm building AI-powered onboarding accelerators with Claude + GitHub + Harness — including a one-shot repo → CI/CD framework with integrated STO, and the IDD onboarding chatbot used internally and by customers.
Krish Mehta.
I ship complete products.
TripOS is a 126-route Next.js 15 app with 689 unit tests and 148 e2e specs. Blackjack Trainer is a pure-function game engine with invariant tests. I take things from blank repo to live URL.
The AI concierge that takes over once your trip starts. 11 agents ship it.
Pure-function engine, optimal-strategy feedback, real card-counter drills.
I ship. TripOS — Next.js 15 + Postgres + Prisma + Stripe + Auth.js v5 + S3 + OpenAI, 126 API routes, 689 unit tests, 148 e2e specs. Blackjack Trainer — React 19, pure-function game engine with invariant tests.
At PayPal I was Software Engineer for Artifactory / GitHub / Jira across 20,000 developers and 2M daily ops — wrote and shipped code, owned tooling, drove rollouts.
Trinity College Dublin, engineering. Bengaluru since.
- RoleSoftware Engineer · @ Harness (Impl.)
- PriorSoftware Engineer @ PayPal
- EduTrinity College Dublin · Engineering
- BasedBengaluru · IST
- StatusAvailable 2026
╔══════════════════════════════════════════════════════════════════╗ ║ KRISH MEHTA · PLATFORM ENGINEER · BENGALURU · 2026 ║ ╚══════════════════════════════════════════════════════════════════╝
EMPLOYER="Harness"
FOCUS="CI/CD · STO · IaCM · IDP · SCS"
PRIOR="Software Engineer @ PayPal — 20k devs, 2M daily ops"
Platform engineer.
Moved PayPal's Artifactory from bare-metal + 5 VMs to a Kubernetes cluster — 40% less downtime. Built Terraform automation cutting 40% of manual repo work. Now owning enterprise CI/CD, STO, IaCM, IDP and SCS rollouts at Harness.
> Bengaluru / remote / hybrid · IST
# ─── system.metrics ──────────────────────────────
# ─── kubectl get deployments --all-namespaces ────
| name | status | scope | impact |
|---|---|---|---|
| artifactory-k8s | Running | 2 bare-metal + 5 VMs → k8s cluster | -40% downtime |
| repo-automation | Running | Terraform + Python · 100+ req/mo | -40% manual |
| emu-migration | Complete | 15,000+ GitHub repos → EMU | -30% mgmt time |
| xray-e2e | Running | JFrog Xray · 250M+ artifacts | 5k+ CVEs mitigated |
| datadog-metrics | Running | 30+ custom metrics · 10+ teams | -20% detect time |
| artifactory-cleanup | Cron | Automated retention + GC | -$60K / mo GCP |
# ─── tail -f /var/log/war-stories ────────────────
Inherited 2 bare-metal hosts + 5 VMs running production Artifactory for 20,000 developers. Sized a k8s cluster, designed a zero-downtime cutover with a read-only window, shadowed traffic for 48h, flipped DNS. Result: -40% downtime, same hardware budget.
Artifactory storage growing unbounded. Wrote a retention policy + GC cron that respected legal/compliance holds. Dry-ran against prod for a week before cutover. Result: -$60K/month GCP, zero accidental deletions.
Migrating 15,000 GitHub repos to Enterprise Managed Users — every dependent CI/CD reference would break. Wrote a Python + Terraform pipeline that rewrote refs, validated, and rolled back per-repo on failure. Result: -30% ongoing repo-management time.
# ─── platform.pipeline ───────────────────────────
# ─── apt list --installed ────────────────────────
# ─── cat README.md ───────────────────────────────
I do platform work. PayPal Artifactory: bare-metal + 5 VMs → Kubernetes cluster, 40% downtime cut. Terraform + Python automation for repo sunsetting and user provisioning — 100+ monthly requests, 40% manual work eliminated.
15,000+ GitHub repos migrated to EMU. 30+ custom Datadog metrics with 20% detection-time improvement. JFrog Xray covering 250M+ artifacts; 5,000+ vulnerabilities mitigated. Artifactory cleanup automation: $60K/month in GCP savings.
Now at Harness, owning enterprise CI/CD, STO, IaCM, IDP and SCS rollouts — integrating with customer code repos and artifact registries.
# ─── send.message ────────────────────────────────
Making enterprise adoption stick.
I work as a trusted technical advisor across five-plus enterprise rollouts spanning CI/CD, security testing, infrastructure as code, internal developer portals, and software supply chain. The job is to take Harness from contract to actually-adopted — architecture conversations, integration design, onboarding strategy, and the kind of AI-powered accelerators that compress the timeline.
I built three onboarding accelerators with Claude, GitHub and Harness — including a one-shot framework that takes a repository all the way to a production-ready CI/CD pipeline with integrated STO. I also built the IDD onboarding chatbot used internally and by customers.
Before this, a year and a half as Software Engineer at PayPal for Artifactory, GitHub and Jira — 20,000 developers and roughly two million daily operations. Strong measurable outcomes (see By the Numbers below).
“Quietly turns a contract into adoption.”
One-shot repo → CI/CD framework.
AI-powered onboarding accelerator built with Claude + GitHub + Harness. Takes a customer repository and produces a production-ready CI/CD pipeline with integrated security testing (STO). Compresses what was a two-week kickoff into a single working session.
IDD onboarding chatbot.
AI assistant for the Implementation Delivery Development process — used by Harness internal teams and customer-side engineers during enablement. Reduces the “where do I find…” tax that bogs every multi-module rollout.
Reusable expert-skill workflows.
Modular Claude/Copilot-style assistant skills that Implementation Engineers can compose for specific customer guidance — accelerating platform adoption and onboarding across multiple engagement types.
Global Financial Services
Multi-module rollout with custom integration to enterprise artifact registries.
Global HR-Tech Platform
Production CD pipelines integrated with their existing repo strategy and release cadence.
Global Marketplace Platform
Four-module transformation. End-to-end from contract to adoption across distributed teams.
Write to the editor.
krishmehta77@gmail.comIndia Standard Time. Async friendly. Direct line preferred for substantive inquiries — I read everything and reply to most within a day.