EduAstra — AI Browser Agent

Your browser agent: ask about any page, draft mails, set up meetings, act across pages on your command. Never submits for you.

Automation — the highest-leverage unbuilt thing: The agent reacts to commands. The leap is scheduled, autonomous automation — EduAstra acting without you asking. "Every morning check the TCS careers page and tell me if NQT registration opened." "When my GST ARN status changes, ping me." This is the difference between a tool you open and an engine that works while you sleep. Technically: a cron + the agent + WhatsApp. It's also the most dangerous to ship before pilot data tells you what people actually want watched. Students — the one that compounds: Proof auto-capture is the killer. Right now a student manually pastes proof into Sprint. Wire the extension to watch: it sees the submitted application, the LeetCode green tick, the mock-test score page → one click sends it to Sprint as verified evidence. This fuses the extension into the execution loop and makes verification effortless — the single biggest retention lever you have. After that: squad layer (the cohort of 8-12, leaderboard of verified proofs — quitting gets socially expensive, which is your whole thesis) and captain dashboard (stall detection → the human call). Vellum — steal exactly one thing: Their consolidation memory (the 4-hour pass that promotes facts, the knowledge graph). You have flat facts + loops today; their graph is genuinely better at long-horizon recall. But it's weeks of work for a marginal user-visible gain at your scale. Don't copy it. What you should copy is their positioning confidence — they raised $25M with no model by owning one crisp idea. Your crisp idea is sharper than theirs: they remember you, you make people finish. Lean harder into that than into matching their memory tech.