Paid, but not posting: an LLM that writes the job post
Employers were buying subscriptions and starting job posts, but many never finished them. I found the form was the problem — so we let them speak or type the role in Hindi or English and had Claude write the post for them to review.
01The signal
Employers pay for a subscription to post jobs on Timbuckdo. Subscription sales were healthy, and employers were starting job posts. But many of those posts were never completed.
That's a bad outcome for everyone: the employer paid but got no value, students had fewer jobs to apply for, and an unused subscription is unlikely to be renewed.
02Root cause
Digging into who was dropping off, most were small business owners and small enterprises. They knew what work they needed done, but struggled to turn it into a proper job post — especially the skills section and the other details a good listing needs.
Insight: employers had the demand and had already paid. The blocker was translating "I need someone to do this" into the structured fields our form asked for.
03The idea
Instead of asking employers to fill a form, let them speak or type the role in their own words, and have an LLM produce the complete job post: title, description, skills, tags, pay range and recommended skill set. The employer then reviews it, edits anything they don't like, and posts.
Key product decisions
- Their language, their way. Employers can speak or type the role in Hindi or English. Support for more Indian languages is in progress.
- Fill the whole post, not just skills. Skills were the worst step, but every field was friction for these users.
- AI drafts, the employer decides. Every field, including the suggested title, can be edited. Nothing is posted until the employer has reviewed the draft.
- Built into the existing flow. The tech team integrated the LLM directly into the job-posting portal, so employers didn't need to learn anything new.
04The solution
05Example
Illustrative example, written for this portfolio to show how the feature works. Not real platform data.
"Mujhe apni kapdon ki dukaan ke liye weekend pe do ladke-ladkiyan chahiye jo customers ko handle kar sakein aur billing mein thodi help karein. Dukaan Lajpat Nagar mein hai."
"I need two people for my clothing store on weekends who can handle customers and help a bit with billing. The store is in Lajpat Nagar."
- Title
- Weekend Retail Sales Assistant
- Description
- Help customers find the right products, keep the store organised and support the billing counter at a clothing store in Lajpat Nagar, Delhi. Weekends only; 2 openings.
- Skills
- Customer handling · Basic billing · Hindi & basic English communication
- Tags
- Retail · Weekend · Part-time · Delhi
- Pay range
- ₹500–700 per day
Employer reviews, edits anything that doesn't fit, and posts.
06Results
Measured on the analytics dashboard over the 2 months after launch:
- Job posts grew 24% in 2 months.
- More job posts led to more job applications from students.
- More employers bought monthly subscriptions to post jobs.
- Platform revenue grew about 8%, driven by more job posts and more jobs completed.
07What I learned
- Paid but unused is an early churn warning. Healthy subscription sales hid the fact that employers weren't getting value.
- Forms are written for the platform, not the user. Small business owners think in tasks, not in "skills" and "tags".
- Use AI to remove work, then keep the human in charge. The draft does the hard part; the review step keeps employers in control of what's posted.
Want the detail? Read the product requirements for this feature.