Job Tracker — Full Project Documentation

Overview

AI-powered job search assistant for Michael Sounhein. Scrapes IT job listings from Indeed/LinkedIn/Glassdoor via JSearch API, scores them against Michael's resume using GLM-5.1, presents them in a Linear-inspired dark UI, and learns from his rejection/apply behavior to improve future matches.

URL: http://192.168.87.30:5100 (LAN) | http://localhost:5100 (local)
Project path: ~/.openclaw/workspace/projects/job-tracker/
Git: commits in main workspace repo

---

Architecture

┌─────────────────────────────────────────────────┐
│  React Frontend (Vite → dist/)                   │
│  Served by Flask as static files on port 5100    │
│  Pages: Dashboard, NewJobs, AllJobs, JobDetail,  │
│         AppliedJobs, Profile, Settings           │
└────────────────────┬────────────────────────────┘
                     │ /api/*
┌────────────────────▼────────────────────────────┐
│  Flask Backend (app.py)                          │
│  - REST API for jobs, search terms, profile      │
│  - Chat API with GLM tool calling                │
│  - Triggers scraper on demand                    │
└────────┬───────────┬───────────┬────────────────┘
         │           │           │
    ┌────▼───┐  ┌────▼───┐  ┌───▼──────┐
    │Scraper │  │Scorer  │  │ Learner  │
    │(JSearch│  │(GLM-5.1│  │(behavior │
    │ API)   │  │ via GW)│  │ analysis)│
    └────────┘  └────────┘  └──────────┘
         │           │           │
    ┌────▼───────────▼───────────▼────┐
    │  SQLite: data/jobs.db           │
    │  Tables: jobs, search_terms,    │
    │  profiles, rejection_patterns,  │
    │  blocked_companies, chat_messages│
    └─────────────────────────────────┘

---

File Map

projects/job-tracker/
├── .env                          # Environment vars (API keys, config)
├── .venv/                        # Python 3.12 venv
├── data/
│   ├── jobs.db                   # SQLite database
│   └── last_job_summary.txt      # Last scraper run summary
├── backend/
│   ├── app.py                    # Flask API server (all routes)
│   ├── config.py                 # Configuration class
│   ├── database.py               # SQLAlchemy engine, session, init
│   ├── models.py                 # Job, SearchTerm, Profile, RejectionPattern, BlockedCompany
│   ├── models_chat.py            # ChatMessage model
│   ├── scraper.py                # JSearch API scraper
│   ├── scorer.py                 # GLM-5.1 job scoring via OpenClaw gateway
│   ├── learner.py                # Behavioral analysis (reads rejections → patterns)
│   ├── chat_tools.py             # 6 tools the AI can call during chat
│   ├── notifier.py               # Telegram notifications
│   └── requirements.txt          # Python dependencies
├── frontend/
│   ├── src/
│   │   ├── App.jsx               # Router, sidebar, layout, collapsible
│   │   ├── api/client.js         # Axios API client
│   │   └── pages/
│   │       ├── Dashboard.jsx     # Stats overview
│   │       ├── NewJobs.jsx       # New job list with quick triage (🎯❌)
│   │       ├── AllJobs.jsx       # All jobs with filters
│   │       ├── AppliedJobs.jsx   # Applied/Will Apply jobs
│   │       ├── JobDetail.jsx     # Split view: iframe + AI chat
│   │       ├── Profile.jsx       # Resume & search preferences
│   │       └── Settings.jsx      # Settings page
│   └── dist/                     # Built static files (served by Flask)
├── scripts/
│   ├── init_db.py                # Create all tables
│   ├── run_scraper.py            # CLI scraper trigger
│   └── seed_resume.py            # Seed resume from file
└── DESIGN.md                     # UI design system (Linear-inspired)

---

Database Schema

jobs

Column	Type	Notes
id	INTEGER PK	Auto-increment
title	TEXT	Job title
company	TEXT	Company name
location	TEXT	Job location
remote	BOOLEAN	Remote eligible
url	TEXT	Link to original listing
description	TEXT	Full job description
salary_min / salary_max	INTEGER	Annual salary range
source	TEXT	Indeed/LinkedIn/Glassdoor
source_id	TEXT	Unique ID from source
match_score	INTEGER	0-100 from GLM scoring
match_reasoning	TEXT	Why GLM gave that score
status	TEXT	new/will_apply/applied/interview/offer/rejected
rejection_reason	TEXT	Why rejected (for AI learning)
apply_note	TEXT	Why applying (for AI learning)
applied_date	DATETIME	When applied
found_date	DATETIME	When scraped

search_terms

Column	Type	Notes
id	INTEGER PK
term	TEXT	Search query string
source	TEXT	'user' or 'ai'
active	BOOLEAN	Enabled for scraping

profiles

Column	Type	Notes
id	INTEGER PK
resume_text	TEXT	Full resume text
location	TEXT	Default: "Appleton, WI"
max_distance	INTEGER	Default: 25 miles
preferences	TEXT	JSON blob

rejection_patterns

Column	Type	Notes
id	INTEGER PK
pattern	TEXT	e.g., "contract-only roles"
source	TEXT	'ai' or 'user'
active	BOOLEAN

blocked_companies

Column	Type	Notes
id	INTEGER PK
name	TEXT	Company name
reason	TEXT	Why blocked

chat_messages

Column	Type	Notes
id	INTEGER PK
job_id	INTEGER FK → jobs(id)
role	TEXT	'user' or 'assistant'
content	TEXT	Message text
created_at	DATETIME

---

API Endpoints

Jobs

• GET /api/stats — Dashboard stats (counts by status, top match)
• GET /api/jobs?status=new&sort=match_score — List jobs (filter by status, sort)
• GET /api/jobs/<id> — Single job details
• PUT /api/jobs/<id>/status — Update status {status, note}
• POST /api/scrape — Trigger scraper run

Chat (AI Agent)

• GET /api/jobs/<id>/chat — Get chat history for a job
• POST /api/jobs/<id>/chat — Send message, get AI response with tool calling
  • Request: {message: "..."}
  • Response: {id, role: "assistant", content: "...", created_at: "..."}

Profile & Config

• GET/PUT /api/profile — Resume text, location, preferences
• GET/POST/DELETE /api/search-terms — Manage search terms
• GET/POST/DELETE /api/blocked-companies — Manage blocklist

Frontend

• GET /* — Serves React SPA from frontend/dist/

---

AI Chat Tool Calling

The chat agent has 6 tools it can call during conversation:

1. update_resume — Rewrites resume text. AI crafts polished bullets.
2. add_search_term — Adds job title keywords to scraper.
3. block_company — Blocks a company from future results.
4. add_rejection_pattern — Adds a pattern to penalize similar jobs.
5. update_job_status — Changes job status from chat.
6. update_profile — Updates location, distance, preferences.

Tools use OpenAI-compatible function calling format. The backend loops up to 5 iterations to handle chained tool calls.

---

Scoring Pipeline

1. Scraper fetches jobs from JSearch API (Indeed/LinkedIn/Glassdoor)
2. Learner reads rejection patterns and builds behavioral context
3. Scorer sends each new job + resume + behavioral context to GLM-5.1
4. GLM returns match_score (0-100) and match_reasoning
5. Match reasoning is stored and shown as AI's opening message in chat

---

Learning Pipeline

Every scraper run:

1. GLM reads recent rejection reasons + apply notes
2. Extracts patterns: "rejects contract roles", "likes remote + MECM"
3. Stores patterns in rejection_patterns table
4. Adds/removes search terms based on patterns
5. Injects behavioral context into future scoring prompts

---

Cron Jobs (Automated Scans)

ID	Time (CDT)	Schedule
1ac6b0f7-...	7:00 AM	Daily morning scan
98e83326-...	12:00 PM	Daily midday scan

Each triggers scripts/run_scraper.py → scrapes, learns, scores, sends Telegram summary.

---

Systemd Service

• Unit: ~/.config/systemd/user/job-tracker.service
• Working dir: ~/.../projects/job-tracker/backend
• Command: .venv/bin/gunicorn app:app --bind 0.0.0.0:5100 --workers 2
• Restart: systemctl --user restart job-tracker
• Logs: journalctl --user -u job-tracker -f

---

Key Integrations

Service	Purpose	Auth
JSearch RapidAPI	Job scraping	Key: cea608bc...
OpenClaw Gateway	GLM-5.1 proxy	localhost:18789, token 172663...
Telegram	Notifications	Chat -1003791430087, topic 1965

---

Design Decisions

• Four Visual Themes & Custom Hexagon Background — Features an interactive Theme Switcher (sidebar, mobile, settings) with custom SVG background pattern support across:
  • Sleek Arctic (Light): Clean, light slate/arctic aesthetic with subtle grey/blue borders.
  • Cyberpunk Midnight (Dark): Sleek dark mode with vibrant neon purple/cyan accents.
  • Warm Honeycomb (BusyBee Light): Warm parchment/beige background with deep charcoal text and honey-gold/amber accents.
  • Honey Midnight (BusyBee Dark): Cozy dark charcoal-brown background with honey-gold accents.
• Alternating Premium Row Layouts — Clean lists featuring alternating color palettes and left-accent borders computed from the company name, improving readability.
• Single-port deployment — Flask serves React static files (no separate dev server)
• Progressive status buttons — Only show actions valid for current phase
• Auto match reasoning — Scorer's analysis appears as AI's opening message in chat
• Tool-calling agent — AI can modify resume, filters, and job status directly
• Mobile-friendly — Responsive sidebar collapses, full-width content

---

Build & Deploy Commands

# Frontend build
cd frontend && npm run build

# Restart backend
systemctl --user restart job-tracker

# Check logs
journalctl --user -u job-tracker -f

# Manual scraper run
.venv/bin/python scripts/run_scraper.py

# Init DB
.venv/bin/python scripts/init_db.py