Designing Your Job Search Agent
Most job platforms ask you to upload a résumé. This one asks a deeper question:
“How do you want to evaluate jobs?”
Instead of parsing your resume, this system invites you to design your agent — guiding its logic and outputs through editable templates.
🧭 How You Guide the AI
There are two tools at your disposal:
1. 🧠 Prompt Editor (Jinja2 Template)
This is your philosophy of fit. Use it to:
- Write out what matters to you: mission, tech stack, flexibility, salary...
- Ask the assistant to evaluate soft signals: culture, red flags, alignment
- Inject your voice and framing into the analysis
Example:
You are a job search assistant helping a user who wants:
- Remote work in a non-profit
- Focus on data policy and ethical AI
- Avoids defense contractors and VC-backed startups
Rate each job from 1 to 10 for this fit.
Return a JSON object with fields: title, company, location, score, summary, concerns.
The model will use this to interpret each job post and return structured outputs.
2. 📄 Schema Editor (JSON Output Format)
This defines what fields you want in the final table — what to extract.
You can ask the model to return:
- Booleans (e.g.,
is_remote) - Strings (
industry,mission_summary) - Numbers (
salary_estimate,fit_score) - Lists (
required_skills)
Example schema:
{
"title": "string",
"company": "string",
"location": "string",
"is_remote": "boolean",
"salary_estimate": "number",
"red_flags": "string"
}
This schema is enforced during the LLM call. If it doesn’t match, the row may be skipped.
🎯 Why This Is Powerful
- 🧠 You define the logic — not a black-box recommender
- ✏️ It’s editable mid-session (via tabs in the UI)
- 📊 Your output is structured for filtering, scoring, and export
Unlike resume-based tools, this system doesn’t guess who you are — it follows what you tell it.
👤 Example Users
Here’s how different users might shape their agent:
| User | Prompt Emphasis | Schema Fields |
|---|---|---|
| NGO analyst | “Impact-focused orgs” | mission_match, concerns |
| Frontend developer | “React, Typescript, remote-first” | stack_keywords, location, salary_range |
| Academic pivoting | “Policy roles for PhD in social science” | relevance_score, team_description |
You can design your own pattern — minimal or rich, relaxed or strict.
🔄 Dynamic Guidance
Every time you:
- Edit the prompt template
- Adjust the JSON schema
- Rerun a query...
...you’re redefining how the system works for you.
This makes it feel less like a form — and more like building your own job search assistant.
🧠 Summary
Instead of uploading a resume, you teach the system how to think. It becomes your proxy — shaped by your logic, your preferences, and your structure.
Want to see how this affects job scoring? Next: How Job Scoring and Ranking Works.