AI Engineering5 min
RAG for job search
RAG over job listings has a problem most RAG systems do not: the documents are written by strangers, and some of them are trying to manipulate you.
Read moreCareer guides thatrespect your time
Updated
Page 6 of 11 · 205 guides in total
AI Engineering5 min
A job recommender, end to end
The interesting design questions are not about the algorithm. They are about what you log, how often you recompute, and what you show someone who signed up four minutes ago.
Read moreAI Engineering5 min
Building a job recommendation engine
The hard part of a job recommender is not the model. It is that the inventory expires, the user has no history, and the feedback you collect is biased by what you showed.
Read moreAI Engineering4 min
Designing a relevance score
A relevance score is a promise to the user. If 80% does not mean something specific and consistent, the number is decoration.
Read moreAI Engineering5 min
Building semantic job search
Semantic search feels magical in a demo and disappoints in production for reasons that are entirely predictable — and mostly avoidable.
Read moreAI Engineering5 min
Evaluating a matching model
Most teams cannot tell whether a change improved their matching. That is not a metrics problem — it is that nobody built the evaluation set first.
Read moreAI Engineering5 min
LLM re-ranking for job listings
A model ranks job listings well and inconsistently. Both halves of that sentence determine how you should use it.
Read moreAI Engineering5 min
Building a career assistant on RAG
A career assistant is a retrieval problem wearing a chat interface. Most of the work is deciding what it is allowed to see.
Read moreAI Engineering5 min
A job agent over MCP
MCP makes your job tools portable across clients. It also means a component you did not write is deciding what your agent can see.
Read moreAI Engineering5 min
Agents that run for weeks
A job search is not a request. It runs for weeks, across restarts and deploys, and every assumption that suits a request breaks.
Read moreAI Engineering5 min
Multi-agent job search
Multiple agents are worth it when the subtasks genuinely differ. They are not a way to make one unreliable agent reliable by cloning it.
Read moreAI Engineering5 min
Deciding what is worth applying to
Scoring a job is easy. Deciding whether it is worth the ninety minutes a good application costs is a different question entirely.
Read moreAI Engineering5 min
Learning what someone actually wants
The risk is not failing to learn. It is learning too fast from too little and locking someone into a version of themselves from three weeks ago.
Read moreAI Engineering5 min
Tool design for a job search agent
An agent is mostly its tools. Most agent failures that look like reasoning problems are tool design problems wearing a disguise.
Read moreAI Engineering5 min
A job search agent as a graph
The reason to model an agent as a graph is not elegance. It is that a job search runs for days and something will crash in the middle.
Read moreAI Engineering5 min
Memory for a job search agent
Memory is not a bigger context window. It is deciding what is worth keeping, and being willing to throw the rest away.
Read moreAI Engineering5 min
One agent or several?
Most teams reach for multiple agents too early, for a problem that was really a tool design problem.
Read moreAI Engineering5 min
Automating Greenhouse applications
The form is the easy part. What decides whether the application is any good is the custom questions the employer added underneath it.
Read moreAI Engineering5 min
Automating Lever applications
The parse-and-prefill step is the whole game here: it saves the candidate effort and quietly introduces the errors a recruiter will read.
Read moreAI Engineering6 min
Automating Workday applications
Workday is where most application automation quietly stops working, and the reasons are structural rather than technical.
Read morePut the advice to work
Reading about ATS screening helps. Running your actual resume through one helps more.
New guides added regularly