Vennbee co-founder · Forward Deployed Engineer (Full Stack) · Aryan Tah · Riverside, CA
I build systems real people run on.
Built, then measured
02 · Swarm rebuild · 39 services → 14 · 10s → 6ms
I design system architecture, then optimize it for latency and throughput.
Starts once it's fully in view · hover any node
03 · Work
Where the systems came from.
I build the search
behind Vennbee.
search
Each subsystem I own, traced through the backend. Pick one to hold it.
vennbee / backend
Vennbee · Forager search
Search that reads like Google, inside a nested social network.
One sentence, like “friends of friends in SWE Internship 2027 and ACM at UCR who placed at Cutie Hack and know Python”, and Forager searches people, spaces, posts and threads inside the communities you're in. A model turns the sentence into a plan. Code checks every limit before anything runs.
- Nine kinds of facts to filter on (skills, roles, orgs, schools, places, languages and more), each past, present or future.
- Scopes and edges: one space or several at once, people you vouch for, an author, “people like @maria”.
- Keyword, semantic and facts lanes, fused, then reranked by how close each result is to you in the graph.
- Too specific? A relaxation ladder drops one constraint at a time and says which, instead of returning nothing.
- Runs on its own Postgres kept current by logical replication. Every result is re-checked against permissions before you see it.
- 84 labelled queries gate every parser change in CI.
Plan
Pipeline
Every plan uses fields Forager's plan schema really has. Two queries are from Vennbee's search eval set; the rest combine its features.
04 · Case studies
How the hard parts got solved.
05 · Projects
Things I built to find out.
06 · Background