Send four people to a conference. Make sure they donโt all learn the same thing.
I built ๐๐๐๐๐๐๐๐ for the
@convex All Gas Hackathon. Parallel turns a public conference agenda into one coordinated team plan, repairs that plan when real life changes and turns everyoneโs takeaway into one source backed brief.
We tested Parallel on the public ViVE 2026 agenda (unofficial).
Firecrawl pulled the agenda, OpenAI scored 9 sessions against 4 team goals, and Parallelโs deterministic optimizer split the team to maximize coverage.
Then we broke the plan.
A teammate emailed that he couldnโt make a session. Parallel verified the exact sentence and marked the plan stale. The lead hit Repair, and another teammate replied YES to cover.
Coverage: 90.5 โ 88.8 โ 90.5.
Takeaways came back through email, get attached to the right sessions, and OpenAI turns them into a source-backed team brief before AgentMail delivers it.
We also built /judges to attack four production guards: duplicate webhook, stale write, invented AI quote and repeated send.
Each Sponsor powers a different part of Parallelsโa workflow:
@convex โ the coordination layer: real time state,plan revisions, repair logic,workflows, webhooks, guardrails and hosting.
@firecrawl - the source layer: scrapes public agenda scraping, stores provenance and detects when published schedule changes.
@OpenAI โ the understanding layer: structure agendas, scores session against team goals, interpretes replies and writes the final brief.
@agentmail โ the human layer: plans, cover requests, replies + delivery
One workflow, four distinct jobs .Check out the product hereโฌ๏ธ
Parallel App:
joyous-akita-768.convex.site
Production Run:
joyous-akita-768.convex.siteโฆ
GitHub Repo:
github.com/Enoch208/parallel
๐ฏ๐๐๐ ๐๐ ๐ 3 ๐๐๐๐๐๐ ๐๐๐
๐๐ ๐๐๐๐๐๐๐๐๐๐๐ ๐๐ ๐๐๐ ๐๐๐ ๐๐๐๐
๐๐๐ ๐๐๐๐๐
@waynesutton @jamwt @erquhart_ @mikeysee @nicolapps @max_kelly @haakamaujla @vedantdotrs