Hours a week vanish into mail only you can answer. Build the bot that answers it instead — and can be stopped mid-send.
Most of your week isn't hard, it's repetitive: the same five kinds of email, the same three answers, the same questions from the same people, all of it eating hours you would rather spend on work only you can do. This course builds the thing that takes those hours back — a personal bot on Telegram, connected to your Gmail account, that reads incoming mail, sorts it by sender and intent, clears the junk, drafts replies in your own writing voice, researches questions you throw at it, and, for contacts you approve, answers on your behalf. You ship a small working piece every module: a deployed bot with a spending cap, a memory layer in SQLite, an email classifier, a drafting pipeline, a countdown that shows you every outgoing reply before it leaves, and a kill switch. Two rules run through all of it. A grace window gates anything the bot sends; verify-before-use gates any fact that enters its long-term memory. You finish by running the bot against your real inbox for fourteen straight days and reporting, honestly, what it got wrong.
Built by Lakshya Kumar
We grant free access case-by-case — students, career-switchers, builders on a tight budget. Sign in to send us a note.
Sign in to applyFinished the tasks? Take the prompt to your AI and get tested on it. We copy the prompt and open the app — just paste it in.
A bot nobody can reach is just a script. Get yours online, holding context, and refusing to run past your monthly budget.
Without memory the bot meets you as a stranger every time. Store what it should know — and the rules for what it must forget.
Before a bot can answer your mail it has to understand it. Turn a raw inbox into decisions you can act on and audit later.
One wrongly deleted invoice sinks trust in the whole system. Automate the cleanup with a false-positive log watching your back.
A reply that reads like a chatbot is worse than no reply. Draft in your voice, at the right length — and know when to stay silent.
Approving every draft is just email with extra steps. Let low-stakes replies send themselves — behind a timer you can always stop.
One confident wrong fact in memory poisons every answer after it. Make the bot show its sources and earn your approval first.
To whoever is on the other end, the bot is you. Decide what it may say, what it must escalate, and who it should never answer.
A bot that works on your laptop proves nothing. Run it for two weeks on real work and let the audit log say whether it earned trust.
Complete all modules, then submit the required number of capstone projects. Each must earn a passing rating from an admin reviewer.
Deploy the bot from all 10 modules to a live host (Fly.io / Railway / your own VM). Keep it running for ≥ 14 continuous days handling your real email + Telegram DMs under the grace-window + verify-before-use disciplines. Submit: GitHub repo, deployment URL, delegation matrix, threshold config, audit-log excerpts (30+ decisions), spam false-positive log (5+), research-memory dump (3-5 verified briefs), before/after time audit, cost log, honest failure log (5+).
Paste this into any AI chat. Fill in the bracketed parts with your context — you'll get back a straight answer on whether this belongs on your plate.
I'm taking "Replace Yourself" — a course about building a personal Telegram+email bot that answers on my behalf under two disciplines: a grace window on outbound actions and verify-before-use on knowledge that enters the bot's authoritative memory. Here's my context: 1. My week's shape (routine / judgment / learning): [describe] 2. My messengers of choice (Telegram / Slack / iMessage): [pick] 3. My inbox (Gmail / Outlook / other): [pick] 4. My stack comfort (Node / Python / both): [pick] Given that, answer: - Which recurring tasks in my week are safest to delegate first, and why? - What's my expected LLM + hosting cost per month at moderate usage? - What's the ONE task that a single misfire would burn the most trust — how should it be gated (rung, threshold, escalation)? - What's the smallest working bot I could ship this weekend, and the one metric I'd judge it by?
The reference you live inside for Modules 2, 7, and 9.