Paste a hiring post into Telegram. Get back a one-page resume PDF tailored to that role and a drafted application email — and nothing sends until you approve it. This wizard writes every config file the bot needs.
Runs in your browserKeys never leave this page~10 minutes
00What you are building
A private Telegram bot, running on your machine or your own server. One user: you.
The pipeline
01
Extract
Role, company, contact email pulled from the post
02
Research
Web search on the company, cited sources only
03
Tailor
Picks your 3 best-fit experiences for the role type
04
Compile
LaTeX to a one-page PDF, verified as one page
05
Review
You approve in Telegram, then it sends
What you need before starting
A Telegram account. Free. You will create a bot in Step 1.
An OpenAI API key with credit on it. Every model call runs on your key, so you pay OpenAI directly — roughly a few cents per application.
Your resume, as a PDF or as text you can paste. More than one is better.
An email account you can send from. Gmail with an App Password is the easy path.
Python 3.11+ and git on the machine that will run the bot.
Where your keys go: nowhere. This page is a static file with no backend, no database, and no analytics. Your API key, bot token, and app password are typed in, written into the .env you download, and forgotten when you close the tab — they are never saved to this browser and never sent to any server belonging to this project, because no such server exists. The only outbound calls are from your own browser directly to api.openai.com (on your key, billed to you) and api.telegram.org.
Don't want to trust a hosted page with a key? Clone the repo and open web/index.html from your own disk. Every step works identically offline, and then the file cannot have been swapped underneath you.
01Telegram bot token
The bot is how you talk to the pipeline. You create it by chatting with Telegram's own bot, BotFather.
Create the bot
Open Telegram and search for @BotFather. Open the chat with the one that has a blue verified checkmark.
Send /newbot.
It asks for a display name — anything, e.g. My Job Copilot.
It asks for a username — must be unique and end in bot, e.g. sam_jobcopilot_bot.
It replies with a token that looks like 8123456789:AAH4x.... Paste it below.
Open your new bot's chat and press Start, so it is allowed to message you.
Treat the token like a password. Anyone holding it controls the bot. Never commit it to a public repo — the .gitignore in this project already excludes .env.
Your numeric user ID
The bot answers only this ID. Everyone else gets "this bot is private" — that is what keeps strangers from spending your OpenAI credit.
Easiest: send any message (e.g. hi) to your new bot, then press Detect my ID below.
Or: message @userinfobot on Telegram and it replies with your ID.
02OpenAI API key
One key powers everything: reading the post, searching the web for the company, and writing the resume.
Add credit first if the account is new: Billing. A key with no credit fails with a quota error.
Click Create new secret key, name it job-copilot, and copy it. You cannot view it again after closing the dialog.
Paste it below and press Verify key.
A ChatGPT Plus subscription is not an API key. API usage is billed separately, pay-as-you-go, from the platform account above.
03Your resumes
Add every resume version you have. The model reads them all and builds one deep profile, which is what the bot draws from later.
Why more than one helps: if you have a technical resume and a product resume, together they contain more detail than either alone — and detail is exactly what makes the tailored bullets specific instead of vague.
Upload PDFs
Drop PDF resumes here, or
PDFs are sent to OpenAI as attachments and read directly — no conversion needed on your side, and scanned PDFs work too.
Or paste resume text
Extract
This makes two calls on your key: one to build the structured profile, one to write the long-form "resume brain". Expect 30–90 seconds and a few cents.
04The questions
Same questions for everyone. These are the things a resume cannot tell the bot — and the answers it will treat as facts it may never contradict.
Run the extraction in Step 3 first. These questions come pre-filled from your resumes, and answering them from scratch is much slower.
Q1 · Contact block
Goes into the header of every generated resume, and into the email signature.
Q2 · Education
Q3 · Which experience leads every resume?
Most people have one role that is always their strongest card — a current job, or the most recognisable name. Pin it and it is always included and always first, whatever the posting. Pick "no pin" to let the bot choose all three purely on fit.
Q4 · Your experiences
Check the titles: a title is a fact the bot may never change to flatter a posting, and it is reinstated automatically after generation. Then tag which kinds of role each experience should be considered for.
Q5 · Your projects
Q6 · Achievements
Q7 · Positions of responsibility
Q8 · The email
These paragraphs are drafted from your strongest experience. They stay fixed across applications — only {company} and {role_phrase} get filled in — so edit them until they sound like you.
Q9 · Hard rules about you
Q10 · Roles you are targeting
Used to sanity-check that each bucket you care about has enough experiences tagged to fill three slots.
05Sending email
How the bot delivers the application once you approve it in Telegram.
Copy the 16-character password it shows. Spaces do not matter.
Use the app password, not your Google account password. Google rejects the account password over SMTP — that failure is the single most common setup problem, and the bot's error message will tell you the same thing.
Enable the Gmail API under APIs & Services → Library.
Configure the OAuth consent screen. External is fine; add your own address as a test user.
Credentials → Create credentials → OAuth client ID → type Desktop app.
Download the JSON and save it in the repo as secrets/credentials.json.
Run python -m pipeline.send_email --authorize, approve in the browser window, and it writes secrets/token.json.
Scope requested is gmail.send only — the bot can send mail as you and cannot read your inbox.
06Download and run
Four files. Drop them into your clone of the repo, in the paths shown.
File
Goes to
What it holds
Then run it
First application
Message your bot /start. It should reply with your name in it.
Leave dry run ON for the first few posts — you get the PDF and the drafted email in Telegram, and nothing is sent.
Paste a real hiring post as a plain message.
Check the PDF. Tap 👎 Fix picks when it chose the wrong experiences and tell it what it should have used — that feedback is stored per role type and biases the next run.
When the output looks right, send /dryrun to switch to really sending.
Nothing sends without you. Every run stops on a review card showing the recipient, the subject, and the full body, with Send and Cancel buttons. Dry run additionally blocks sending entirely.
Before you push anything public:.env, secrets/, and your profile/ files are all in .gitignore already. Run git status once and confirm none of them are staged.
Keep it running when your laptop is closed
The repo ships a Dockerfile that installs the LaTeX engine and pre-warms its font bundle. Deploy it to any always-on host (Railway, Render, Fly, a VPS) and set the same variables from your .env in that host's environment panel. For Gmail OAuth on a host with no filesystem, paste the file contents into GMAIL_TOKEN_JSON and GMAIL_CREDENTIALS_JSON instead.