Research
Sweeps configured sources and records coverage, including when a source is unavailable.
Project 05 · AI agents
A job application agent that knows when not to act.
Job Application Agent turns a fragmented search into a controlled workflow: it finds opportunities, evaluates fit, prepares the application, and only calls something a success after the external system confirms it.
Working MVP · local-first · evolving
JOB APPLICATION AGENTLOCAL
01ResearchDONE
02EvaluateDONE
03PrepareDONE
04Request a decisionHUMAN
05ConfirmEXTERNAL PROOF
This is a product demonstration. Roles, companies, documents, messages, and operator history remain private.
The product question
Automation is easy. Trust is the hard part.
An application combines identity, documents, preferences, consent, and irreversible actions. The challenge was not making a bot click faster, but building a system that keeps intent and execution aligned.
How it works
From discovery to evidence, with explicit boundaries.
Evaluate
Separates role fit, opportunity quality, and objective constraints before taking action.
Prepare
Builds materials and answers from verified facts without inventing experience or credentials.
Request a decision
Stops when consent, sensitive information, or human judgment is required.
Confirm
Records an application as submitted only after the external system returns real confirmation.
Trust architecture
The product was designed around what can go wrong.
Privacy by architecture
Product code is versioned; identity, documents, and history stay in separate local state.
Human in control
Explicit permissions define what the agent may execute and where it must stop.
Operational truth
Prepared, blocked, and submitted are different states. The interface never disguises an attempt as an outcome.
Compounding learning
Proven tactics become reusable playbooks while policy and execution remain auditable.
What I learned
A good agent is not the one that does everything alone.
It preserves context, explains state, knows what evidence is missing, and returns the right decision to the right person. Job applications were the use case; the lesson applies to any agent operating in the real world.
← Back to projects