Staff Software Engineer · Agentic Engineering
· 2020 — Present
GURINDER SINGH
Building the nervous system for agentic engineering.
A connected platform that observes work, evolves its own tooling, applies that intelligence to real agents, and learns from validated outcomes.
I build the platforms
USPTO patent approved★BBC documentary featured★90M+ users impacted★Agent observability★Self-evolving agents★AI merchant onboarding★Cited RAG with guardrails★USPTO patent approved★BBC documentary featured★90M+ users impacted★Agent observability★Self-evolving agents★AI merchant onboarding★Cited RAG with guardrails★
Agentic Engine
Two engines. Everything else benefits.
Observe + evolve, then, recalibrate
Vireo observes and remembers. Darwin evolves. MAIA and Devmo turn that into real outcomes — validated in a production-shaped sandbox, with results returning to memory.
01 · Vireoobserves runs, tools, traces, failures, and lessons
→02 · Darwinevolves skills, prompts, subagents, and workflows
→03 · MAIA + Devmomerchant-facing and internal-agent surfaces
OBSERVE · ENGINE
Production · v32026 · ongoing
THE OBSERVABILITY CORE
Vireo
The flight recorder for AI agents
You can't trust an agent you can't see. Vireo captures everything one does — then turns it into memory the next run can use.
Vireo records what an agent actually did — across the CLI, shell hooks, MCP, and skills — and distills those logs into procedural, failure, and semantic memory the next run can draw on. It installs in two minutes and became the team's standard for watching agents in production. With Darwin, it's the engine every other system here runs on.
AI flight recorderWhat it is
CLI · hooks · MCP · skillsWatches
team standardAdopted as
2 minutesInstall
MCP
n8n
OpenTelemetryOpenInference
Python
Swift
EVOLVE · ENGINE
Production + Council pilot2026 · productionized Jun 30
THE SELF-EVOLUTION CORE
Darwin
Skills that improve themselves
If a task has a prompt in, an answer out, and a way to score it, Darwin evolves it — automatically, and opens the PR.
Darwin takes the skills agents run on, tries many variations, keeps the best, and opens a pull request with the upgrade — every attempt saved with its score and reasoning, so each round builds on the last. It plugs into any repo with no fork. This is the loop that quietly improves MAIA and Devmo underneath them.
Self-improves AI skillsWhat it does
scored + preservedEvery attempt
dual-model CouncilReview
any repo, no forkSetup
GEPA
DSPy
Cosmos
Graph DB
GitHub Actions
APPLY · MERCHANT
Live cohort pilots2026 · ongoing
THE MERCHANT-FACING BENEFICIARY
MAIA × Venmo
Cold-start to live in 20 minutes
An ephemeral Claude-CLI agent — built entirely from skills and subagents — that onboards a new merchant end to end.
MAIA is bootstrapped and ephemeral: a Claude-CLI agent assembled from skills and subagents — the same skills Darwin evolves and Vireo observes. It's the merchant-facing payoff of the engine. Over one weekend I grew its test corpus from 332 to 9,513 real scored businesses; onboarding that took ~3 months now runs in about 20 minutes.
~3 mo → 20 minOnboarding
9,513Businesses scored
skills + subagentsBuilt from
Vireo + DarwinRuns on
Claude CLISkills + Subagents
RAGCommonCrawlAWS AthenaD3.js
APPLY · INTERNAL
Production · in flight2026 · ongoing
THE INTERNAL BENEFICIARY
Devmo
Venmo's agent — and its hive mind
MAIA faces merchants; Devmo faces the org. It turns the signals Vireo captures into organizational memory Venmo can act on.
Devmo is the internal counterpart to MAIA. It leans on the same Vireo + Darwin engine, but because it lives inside the org, it can do more: build a hive mind from the organizational insights, lessons, and signals Vireo captures — and evolve far more than skills. It reaches the whole internal surface area of how Venmo's engineers actually work.
the org, not merchantsFaces
an org hive mindBuilds
internal surface areaEvolves
Vireo signalsFed by
VireoDarwinn8nMCP
k8s
Qdrant
Externalization and self-service scale-upVALIDATION
Merchant Sandbox
A production-shaped validation surface for checkout, authentication, release, and partner workflows — tested end-to-end on iOS.
- Helped lead and personally tested merchant sandbox flows end-to-end across the iOS-majority user base.
- Owned iOS identity, domains, entitlements, provisioning, CI, and TestFlight distribution.
- Diagnosed authentication, universal-link, direct-API host, and switchback behavior.
AI Enablement · org-wide leverage
Claude Cosmos
A Claude-CLI harness that lets any Vertex AI model run behind one consistent agent interface — with sanitization, PII scrubbing, and telemetry baked in.
Org-wide proxy
ai-video-toolkit
Turns code and technical work into narrated explainers that keep product launches current — Remotion 4.0 + Google Veo / Chirp3-TTS, runnable from GitHub Actions.
Launch storytelling
n8n
My favorite way to teach others how automation works — and a powerful internal automation tool for the enterprise.
Automation + teaching
Nested macOS
Apple Virtualization gives agents a safe sandbox and parallelizes iOS tests for 2–3× faster runs — a secure runtime substrate for agent experimentation and the iOS test fleet.
Runtime substrate
DX Sounds
MCP server merged into the PayPal AI Hub — 8 themes, 14 tools. Sonic confetti for your dev loop.
In AI Hub
Devmo Prompt Enhancer + Guardrails
Guardrails and prompt enhancer migrated into the canonical Devmo repo. Guardrails ship with ~90 tests including a prompt-injection battery.
ProductionizedStack · 2026
The tools
I reach for first.
PythonClaude
OpenAI
Gemini
Ollama
LangChain
LangGraph
LlamaIndex
CrewAIMCPPythonClaudeOpenAIGeminiOllamaLangChainLangGraphLlamaIndexCrewAIMCPSwift
SwiftUI
Objective-C
ARKit
RoomPlan
CoreML
ActivityKit
WidgetKitSwiftSwiftUIObjective-CARKitRoomPlanCoreMLActivityKitWidgetKit
AWS
GCP
AzureDockerGitHub Actions
Jenkins
Fastlane
VercelAWSGCPAzureDockerGitHub ActionsJenkinsFastlaneVercel
PostgreSQL
Supabase
Cloudflare
MongoDB
RedisQdrant
ChromaDBNeo4jPostgreSQLSupabaseCloudflareMongoDBRedisQdrantChromaDBNeo4jInteractive · RAG demo on real corpus
Ask my brain.
A live retrieval pass over my 2026 work — questions retrieve source chunks, light up systems and exact skills, then answer only from the corpus.
Ask anything.
I'll retrieve relevant chunks from the 2026 corpus and light up the systems and skills behind the answer.
ask to reveal projectsSelect a node
Hover source chips or ask a question to illuminate systems, metrics, and related infrastructure.
Film Reel · The Hackathon Circuit
28 HACKATHONS. TWO FIRSTS.
A BBC DOCUMENTARY.
A decade on the circuit across six countries — as competitor, mentor, and judge. Hover any clip to preview, click to watch the full thing.
Where it's taken me
London 🇬🇧New York 🇺🇸Zurich 🇨🇭Barcelona 🇪🇸Copenhagen 🇩🇰San Francisco 🇺🇸Tokyo 🇯🇵
On camera
1ST PLACE▶
FINALIST▶
AR FITNESS
▶
1ST PLACE
▶
More Hackathons
1st Place
Devcamp NYC @ Google
Talent Management App
Team Build
HackZurich 2023
Legal Lunatics
Open Source
UPC Barcelona
Refugee Crowdsourcing App
Participant
Copenhacks @ Microsoft
Innovation Project
Participant
HackNY @ NYU
Urban Tech Innovation
Judge
Hack Yale (2×)
Hack Yale (2×)
Mentor
PennApps — UPenn
PennApps — UPenn
Conf
try! Swift Tokyo
try! Swift Tokyo
Community and Adventures
Stages, hackathons,
and the next generation.
Talks, winning teams, and mentoring — click any photo to enlarge.
Selected builds · live demos & shipped apps
Curated Showcase.
Six surfaces worth a closer look — landscape homes, hover previews, and the full story behind each build.
SYSTEMS THAT EVOLVE.
Gurinder Singh
Staff Software Engineer
Agentic Engineering
LET'S CONNECT
Interested in collaborating on innovative projects? Get in touch and let's build something legendary together.