Show HN: Good old emails and LLMs for automating job tracking
3 adelowo 5 6/21/2025, 9:37:48 PM
So I spent the last few days building Jobstack. The logic is quite simple. You apply to jobs and you get emails, you trade emails back and forth from interviews, questions and others until the role is either accepted or you are rejected.
Also easy to apply to hundreds of roles and not being to know where you stand easily. With Josbtack, you sign up, get a unique email and forward emails to the url. And it uses LLMs to extract company details , tries to find information online about them and presents that to you.
Every email you forward becomes part of your timeline with the company.
It also tracks rejection, offers from the emails too and gives you a nice stats dashboard amongst others.
Using Gemini 2.5 pro right now. No data stored not in any way. After extraction, it’s discarded. Even “AI chats with the company” aren’t stored
A couple of questions: - How reliable is the rejection/offer detection logic? Are you using simple keyword matching or something more contextual with Gemini? - How do you handle company name normalization across emails? (e.g., "Google LLC" vs "Google Careers")
Also curious: any plans to expose the timeline as a downloadable JSON or API? Would be neat for power users to integrate into personal dashboards.
Great job building this — surprisingly underserved use-case despite how common it is.
I’m using contextual searches with Gemini. Yes it would normalize Google LLC and Google careers as Google.
I didn’t think of that but I can surely get that done next week. That’s a good usecase!
tldr: Nothing is stored in any way, shape or form. Only the status/sentiment of the email content, and company name is being extracted. And others discarded.
The problem isn’t a checklist, it’s keeping it up to date when you apply to tens and 100s of jobs.
It’s a job tracker - whether by email or another SaaS type, you still put in the data regardless. But the policy on my end is not to store anything.
Even the chat feature doesn’t store. You type and it’s fired to the LLM for response
There is no way for a user to verify that claim.