SkillRadar — freelance skill gap analysis engine

The problem

Running a freelance consulting business means constantly deciding: what should I apply to, and what should I learn next? The answers change weekly as market demand shifts. Doing this manually — reading job postings, comparing to your skill set, estimating learning time — is slow and inconsistent.

What I built

Data pipeline:

• Upwork job search results ingested into PostgreSQL via Django management commands
• Skill extraction from job tags + NLP regex pass on descriptions
• 65-skill profile seed covering infra, M365, Python, AI/LLM, networking, security, automation

Gap scoring engine:

• Composite score: demand × adjacency × speed-to-billable
• Demand: frequency of skill in 7-day job window, normalized
• Adjacency: bucket-match against profile categories; Claude Haiku for unknown skills
• Speed: static map (FastAPI = 3 days, Kubernetes = 60 days) with Claude fallback
• Verdicts: exploit_now (≥0.6), learn (0.25–0.6), ignore (<0.25)

Weekly digest:

• Telegram bot message every Monday 8am
• Top 3 exploit-now skills, top 3 to learn, focus category
• Powered by Claude Haiku (~$0.001/unknown skill, 30-day cache)

Proposal tracking:

• Django + Django Ninja REST API
• Upwork MCP auto-logs every submitted proposal
• Stale proposal alerts (>48h no response) via Telegram
• Admin UI at /admin/

Stack

Django 6 · Django Ninja · PostgreSQL · n8n · Claude Haiku · Telegram Bot API · Upwork browser automation (Patchright/CDP)

What I learned

• Vue-controlled sr-only checkboxes need CDP label clicks, not JS event dispatch
• Cloudflare rate limits at 3-4 rapid navigations — 30s inter-request throttle is the safe zone
• GraphQL response interception is more reliable than DOM scraping for balance data

Source code — private repo, available on request.