ANSWER A QUESTION
Search my records, check the best matches, then answer using those passages.
This is the RAG part: find useful source material before answering.UNDER THE HOOD
What’s going
on here?
A real engineering project. A very questionable game-show presentation.
Follow the evidence from job description to approved application.
One career record. Two ways to use it.
My reviewed work experience
What I did, where I did it, and the records supporting it
What I did, where I did it, and the records supporting it
WRITE AN APPLICATION
Read my confirmed career facts alongside the job description, draft the documents, check them, then hand them to me for review.
This is how the applications on this site are created.The walkthrough explains both features. Steps 2 and 3 show how question answering finds useful evidence; application writing uses the full career record and continues with step 4.
CHOOSE A ROUND
ROUND 1 OF 7
Give the system something true to work with
I supply my resume, a record of my past work, and the job description. The career records tell the system what I have actually done; the job description tells it what this employer cares about. Each piece of experience stays connected to the right job or personal project, so work I did at one company does not get credited to another.
For the engineers in the audience
Python parses PDF and UTF-8 Markdown. The career catalog stores confirmed facts, role attribution and source references. Each drafting session saves its own copy of the job description, profile and catalog. Confirmation means I reviewed the source material; it is not independent verification.
Implementation: ingestion.py / resume_drafting.py / tracker/workflow.py
The lifelines have limits.
- A source reference shows where a claim came from. It does not guarantee the claim is correct.
- If my records do not support a skill, the system should flag the gap rather than invent experience.
- If a check fails, the affected operation stops and earlier drafts stay saved. Some invalid model responses get a limited retry; further attempts may use paid API calls.
- The poll remembers a browser without requiring an account. It discourages repeat votes, but someone using another browser can still vote again.
- I use existing AI models rather than training my own. I still review the documents and submit every application myself.