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]
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
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 :)