The technical implementation treated the iteration process like training a model: test the AI's responses, measure the "loss" against what I wanted, backpropagate by adjusting prompts/RAG/CRUD, and repeat.
Happy to answer any questions about the tech stack, the AI architecture, or the broader vision!
Nice job. I've seen a couple of these on HN (Resume-context driven chat interfaces). [1] [2]
I can't speak to whether this could become a realistic standard but you might want to reach out to @jhgaylor who took a stab at trying to build an MCP server around this concept. [3]
Thanks for sharing these examples! I'll definitely reach out to @jhgaylor - would love to learn from their experience with the MCP approach. Appreciate the pointers!
Ha! Even simpler: AI generates perfectly tailored resumes for every job. If you somehow get an interview, build everything with AI in a few hours the night before.
Hey HN! I'm the builder here.
If you're interested in the full story and more details, I also wrote about this on LinkedIn: https://www.linkedin.com/posts/charlie-tianle-cheng-6147a432...
The technical implementation treated the iteration process like training a model: test the AI's responses, measure the "loss" against what I wanted, backpropagate by adjusting prompts/RAG/CRUD, and repeat.
Happy to answer any questions about the tech stack, the AI architecture, or the broader vision!
Nice job. I've seen a couple of these on HN (Resume-context driven chat interfaces). [1] [2]
I can't speak to whether this could become a realistic standard but you might want to reach out to @jhgaylor who took a stab at trying to build an MCP server around this concept. [3]
[1] - https://www.jon-olson.com/resume_ai
[2] - https://replicant.im/alex
[3] - https://news.ycombinator.com/item?id=43891245
Thanks for sharing these examples! I'll definitely reach out to @jhgaylor - would love to learn from their experience with the MCP approach. Appreciate the pointers!
Wow.
> Build a platform where anyone can create their AI twin for genuine matching.
Add a premium tier that deepfakes you into each opening's "AI"-researched ideal candidate.
And a super premium tier to deliver the exclusive best fake for each particular opening.
Once you get traction, offer recruiters a filter that removes the fakes for $$$, but instead deliver just improved fakery.
Let recruiters pay $$$$$ to have their competitors get only fakes.
But be quick, else be beaten to it by Cory Doctorow or Charlie Brooker :)
Ha! Even simpler: AI generates perfectly tailored resumes for every job. If you somehow get an interview, build everything with AI in a few hours the night before.
> AI generates perfectly tailored resumes for every job.
The recruiter's "AI" would spot that all resumes for the same job were identical.
So none would get through ... unless the recruiter knew his client dumb enough to be using an "AI" to choose...