How the AI works
Why a model at all
A rules engine can find the word “Python”. It cannot find distributed-systems experience written as “architected the ingestion layer for a 40 TB data warehouse”, and it cannot judge whether that background is a strong or partial match for a role asking for “large-scale infrastructure”, because that judgment lives in the thousands of job posts and resumes the model has seen.
The harder half is honest tailoring. Rewriting a bullet so it speaks to this role without changing the facts, in the person's own register, is a writing problem. That is the part job seekers pay for and it is the part only a language model does well.
So the model is the engine, and betting the product on one lab would be a mistake. Everything above this layer asks for a task, not a model. This file decides which model answers. Changing a route is one line; adding a lab is one case.
The pipeline
A fast, cheap model reads the job post and works out the role title, seniority level, and domain. This triage pass gives the tailoring step cleaner context and costs almost nothing. Getting classification wrong skews everything downstream.
The balanced model reads the full job post and the full resume together in one pass. It scores fit from 0 to 100, calibrated to the market, and finds the requirements the resume genuinely meets. For each match it returns a verbatim quote from the resume as evidence. Any match whose evidence is not exactly in the resume is dropped before you see it.
The model rewrites the summary and the most relevant bullets to speak to this role, using only what is in the resume. No invented employers, titles, degrees, metrics, or projects. Each tailored bullet traces back to a source line in the real resume. The full tailored resume comes out as clean plain text.
A separate pass because writing and judging are different jobs and may benefit from different settings. The cover letter is specific to this company and role, 200 to 300 words, drawn from the tailoring context so it is consistent with the resume rather than a separate document.
Routing
This is not a diagram somebody drew. It is rendered from the same routing table the tailoring engine reads, so if it is wrong here it is wrong in production.
| Step | Model | Provider | Tier | USD per M tokens |
|---|---|---|---|---|
| Score the fit and tailor the resume honestly | gpt-4.1-mini | openai | balanced | $1.60 |
| Draft the cover letter for this role | gpt-4.1-mini | openai | balanced | $1.60 |
| Build the interview-prep set | gpt-4.1-mini | openai | balanced | $1.60 |
| Work out the role and seniority | gpt-4.1-nano | openai | fast | $0.40 |
Above roughly 24,000 characters, the job post and resume combined, the pass escalates to the frontier model rather than the balanced one. Long or unusual roles need more reasoning capacity and the cost difference is worth it.
If the routed model fails or returns nothing, the call walks a chain of candidates from other tiers and other labs before giving up. One provider having a slow afternoon should not cost a job seeker their tailoring.
Every evidence quote is located in your actual resume before it is shown, and the full response is validated against a schema. A quote that does not appear is dropped and counted. This is the only part of the pipeline the model does not get a vote on.
Candidates
Every model below sits behind the same interface. Adding a lab is one case in one file, which is the entire point of building it this way.
openai · 1,000,000 token context
openai · 1,000,000 token context
openai · 1,000,000 token context
anthropic · 200,000 token context
google · 1,000,000 token context
meta · 128,000 token context
In this deployment the OpenAI and Anthropic adapters are written and the OpenAI one is live. The rest are declared with their real model names and switch on when their key is present. No model is trained on customer resumes or application data, by us or by our providers.