Aditya Singhi
I build things that handle real traffic. Creator of spoo.me, an open-source link infrastructure serving 500K+ daily requests and 140M+ clicks. Built, scaled, and operated solo since I was 17, on about $20 a month.
I spent six months at Emergent (YC S24) on the Deployer team, the one-click deploy system for agent-generated apps: redesigned its MongoDB transfer layer into a stateless control plane, and found the race condition that was burning $700K a year. Mostly 
Python, Go, TypeScript and more Kubernetes than I'd like to admit.
I care about architecture, backward compatibility, and shipping things that don't break when people depend on them. And I'm not backend-only: I ship spoo.me's entire web app too.
github contributions
experience (05)
Deployer team: the one-click deployment system for agent-generated apps. Kubernetes, a dedicated Atlas cluster per app, storage, networking, custom domains, and lifecycle.
- Led the redesign of the MongoDB import/export layer into a stateless control plane: Atlas Federation executes transfers directly to GCS, services only orchestrate, and job status is reconstructed from Atlas telemetry instead of a local state machine. Moves 3 GiB in under a minute, survives service restarts, with no workers, queues, or schedulers to operate.
- Fixed a distributed race condition in pod restart workflows that duplicated ~20K GCP disk snapshots/day, eliminating $700K+/year in infrastructure costs.
- Stopped cascading backup failures that left ~60TB of stuck Kubernetes PVC storage by fixing a concurrency bug behind a self-amplifying retry loop (~5K jobs/hour).
- Shipped an offer feature spanning ~4 services at 11% conversion.
API-first, open-source link management infrastructure (a Bitly/Dub alternative). Built, scaled, and operated solo since I was 17, on a ~$20/month VPS.
- Re-architected the redirect hot path after a botnet saturated the service through a single URL: async Redis Streams fanout plus a Cloudflare Workers/KV edge cache driven by a hotness detector. p95 went from 15s to 5.8ms, ~96% of traffic never reaches origin, and hot links keep serving through a full origin outage.
- Built the edge layer that answers one URL two ways in ~7ms: crawlers get synthesized HTML carrying the link owner's configured meta tags so previews render correctly, humans get a bare 302, all under multi-tenant custom domains.
- Grew it into a platform: a 'Sign in with Spoo' identity layer, multi-tenant custom domains, real-time analytics on MongoDB time-series collections, and abuse defense. 50+ self-hosted forks run in production, some pushing 1M+ redirects of their own.
- Migrated the database off MongoDB Atlas onto self-hosted Percona Server for MongoDB, with PMM query profiling and nightly snapshot backups, keeping the operational guarantees at a fraction of the cost.
projects (13)
skill station
32 skills · 7 projects
the api
yes, this site has one
This portfolio serves its own data over GraphQL. Experience, projects, stars: all queryable. The editor below is live, edit the query and run it.
{
projects(limit: 2, sortBy: STARS_DESC) {
name
stars
}
}// press run
contact
Have something that needs to survive production?
I'm open to internships and new-grad roles (graduating July 2027), freelance infrastructure work, and interesting open-source problems. Use the form or email me directly, I actually reply.