Story of Hirable | AI platform for job seekers
Hi everyone!
I am Oat, founder of https://hirable.fyi with a mission to be an all-in-one job-advancing platform for job seekers.
I am a self-taught frontend engineer that went through a career transition myself 8 years ago. Since then, I've been helping friends and strangers self-transition into tech by setting their mindsets, reviewing their resumes, and guiding them in their job search.
Beginning
3 months ago, with the boom of AI and GPT, I decided to play around with OpenAI's API to build a simple MVP of an AI-assisted resume writer. Upload your resume, paste a job description, and let AI do the rest. Same for analyzing resumes or job descriptions for keyword/hard skills matching etc.
I started writing on Medium about how I started:
https://oat-on-a-boat.medium.com/what-i-am-building-2fbfd08ccbb2
For MVP the app was free and no sign-up was required.
- Next.JS v13 (React.js v18)
- Node.js
- OpenAI's GPT
- Vercel to deploy
Then I wanted to add signup to track usage and growth:
- MongoDB for Database
- Next Auth.js for authentication
- Upstash/Redis for rate limiting
- Google Analytics
- SendGrid for welcome/feature update emails
Challenge 1:
Vercel's serverless architecture only allows 10s API request time before timeout. This shouldn't be a problem for most, but since OpenAI can take time to rewrite or analyze, usually >10s. I had to upgrade to Vercel's premium plan temporarily. The correct solution was to stream the request (both request to OpenAI and then stream that stream to frontend), which would return in <1s and continue until the stream is complete.
Challenge 2:
I wanted to add support for PDF upload to parse their existing resume. This was quite a challenge to build it myself. What I needed was to
- handle PDF upload
- parse that file into its contents
- format that content into the correct interface.
Challenges were that most resume parsing node.js APIs were outdated or had dependencies on canvas module, which for some reason doesn't work well with next.js 13. Additionally, even once we parse the API, we need AI to format the data. I tried asking GPT to format it but it took a long time and made many mistakes. Thankfully I found https://www.superparser.com/ which fitted my need exactly.
Next steps:
I plan to start charging a freemium model monthly subscription. I am exploring https://paddle.com/ but they need 3 months of revenue first before approval (which makes no sense since I need a subscription service to start billing users), so I am exploring Stripe or GumRoad.
Additionally, I plan on adding a career personality assessment functionality for paid users. I will use a version of https://www.big5assessments.com that is affordable.
Costs
Costs with ~500 sign-ups and very few returning users, engagement time of ~5 minutes is manageable. OpenAI charges <$10 per month, Upstash costs <$5 and everything else is free. I will eventually need to pay for SuperParser if my numbers go higher and pay for new services like Big5 Assessment.
Other costs will soon come from advertising and affiliate/influencer marketing - getting promoted in newsletters and running ads at universities for instance.
Thanks for reading!
