Career Studio

v0.42.1.1 — A local web app that turns your LinkedIn profile into a four-part career package — generated in parallel by any OpenAI-compatible LLM and streamed live to your browser.

---

What it produces

Output	Description
LinkedIn Update Plan	New headline options, rewritten About section, experience bullet rewrites, top 10 skills to pin, and a profile hygiene checklist
Master CV	ATS-safe, reverse-chronological, with [FILL: …] placeholders for missing data — never invents employers, dates, or certifications
6–12 Month Dev Plan	Tiered certification roadmap, hands-on projects, networking milestones, and curated learning resources
10 Target Roles	Fit scores, gap analysis, and ready-to-paste LinkedIn and Indeed search queries for each role

All four sections generate in parallel. You watch them stream in live, tab by tab.

---

Quick start

Requirements: Python 3.8+ · Internet access (for the LLM API and pdf.js CDN)

Mac / Linux

cd career_studio
./start.sh

Windows

Double-click start.bat, or in a terminal:

cd career_studio
start.bat

The script installs dependencies, then opens http://127.0.0.1:8000 in your browser automatically.

---

Setup

1. Open the ⚙ API Settings panel in the app
2. Choose a provider preset — this fills in the Base URL automatically
3. Enter your API key
4. Enter a model name

Supported providers

Provider	Base URL	Suggested model
Anthropic	https://api.anthropic.com/v1	claude-haiku-4-5-20251001 · claude-sonnet-4-6
OpenAI	https://api.openai.com/v1	gpt-4o-mini · gpt-4o
Ollama (local)	http://localhost:11434/v1	llama3.2 (or any pulled model)
Any OpenAI-compatible endpoint	custom	—

For Ollama, enter none as the API key. No account or billing needed.

Your API key is sent directly to the provider with each request and is never stored on any server.

---

Inputs

Field	Required	Notes
LinkedIn profile content	Yes	Paste text, or upload .pdf, .txt, .md, .html, .rtf
Personal website / portfolio URL	No	Adds tone and brand context
GitHub URL	No	Engineering signal for the CV and role targeting
Existing CV / résumé	No	Improves continuity of the generated CV
Target job titles	Recommended	Drives all four outputs
Target industries	Recommended	Drives the jobs section
Geography / timeline / constraints	Optional	Affects compensation tone, work-auth notes

PDF uploads use pdf.js as the primary extractor, with a built-in hand-rolled fallback for edge cases. Scanned / image-only PDFs cannot be extracted — paste the text manually in that case.

---

Exports

From the Results panel, download each section or all four at once as:

• Markdown (.md) — raw text, one file
• PDF — opens a print-preview window; choose Save as PDF as the destination
• Word (.doc) — Microsoft Office HTML format; opens cleanly in Word, Pages, Google Docs, and LibreOffice

---

Distribution

The career_studio/ folder is self-contained. Zip it and share it:

zip -r career_studio.zip career_studio/ --exclude "*.pyc" --exclude "*/__pycache__/*"

Recipients unzip, run the start script, and the app is ready. No build step, no Docker, no cloud account required beyond their own LLM API key.

---

File structure

career_studio/
├── server.py               # FastAPI backend — SSE proxy to the LLM API
├── requirements.txt        # openai, fastapi, uvicorn, httpx
├── start.sh                # Mac/Linux launcher
├── start.bat               # Windows launcher
├── public/
│   ├── index.html          # UI shell
│   ├── style.css           # Design tokens, layout, responsive grid
│   └── app.js              # SSE consumer, PDF extractor, Markdown renderer, export
└── data/
    ├── job-sources.json    # 9 curated job-search source categories
    └── sample-profile.txt  # Sample LinkedIn profile for the "Load sample" button

---

Changelog

v0.42.1.1

• Added Refine section — follow-up chat against the generated outputs; supports multi-turn conversation, streaming responses, and full Markdown rendering
• Fixed start scripts (start.sh / start.bat) not finding requirements.txt when launched via double-click or a full path
• Prompt size guarded: follow-up context capped at 10 KB per section / 50 KB total to stay within model limits

v0.42

• Initial public release

---

Security notes

• The server binds to 127.0.0.1 only — not reachable from other machines on your network
• No shutdown endpoint — stop the server with Ctrl+C
• All request fields are validated server-side with strict length caps (LinkedIn / CV ≤ 30 KB, other fields ≤ 2 KB)
• The Markdown renderer sanitises all output and blocks non-http(s):// link schemes
• Your API key lives in browser localStorage only — it is never written to disk by the app

---

License

MIT