尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證NovaWerk Lab, the open workshop of NovaWerk AI: practical AI agents for real work, built in public by Hardik Kathiriya and explained simply.
翻譯準備中
取自儲存庫 README
正在載入完整 README。
檢視完整 README ↗The open workshop of NovaWerk AI, built by Hardik Kathiriya.
🌐 Visit the lab's website: watch each tool run, step by step
Roadmap · See it in action · Start here · Skills covered · How each project is built · Follow along
I'm Hardik, moving into AI engineering from Germany 🇩🇪. My Master's thesis is on Agentic AI and knowledge management.
NovaWerk Lab is where I learn by shipping. It's the open workshop of NovaWerk AI, the AI brand I'm building. I build one small AI tool at a time, and each one solves a real problem. Then I write down exactly how I built it, so you can learn from it, copy it, or build your own.
Every project is:
| 🧰 7 tools live | 🧪 467 checks, all passing | 💶 €0 to run | 🔒 Private by design |
|---|---|---|---|
| shipped between 5 and 17 October 2026 | 416 automatic checks, plus 51 quality checks with the real AI | free models in Ollama, no account and no API key | the AI runs on your laptop, so your CV, letters and notes never leave it |
The idea behind every tool: the AI reads and writes, plain Python checks its work, and you make the decisions. Small AI models make mistakes, so the AI never gets the last word.
Built with Python · Ollama (Gemma 4, EmbeddingGemma) · ChromaDB · Streamlit · pydantic · pytest · openpyxl
| # | Project | What it does | AI skill it teaches | For | Status |
|---|---|---|---|---|---|
| 01 | Job Application Tracker | Excel / Google Sheets job-search CRM with follow-ups, pipeline, analytics and a live dashboard | Automation, data modelling, testing | 🎓 | ✅ Live |
| 02 | Job-Fit Checker | Paste a job ad and your CV. It gives a Match %, your missing skills and a "Why me?" sentence. Runs on your own computer, free | Using AI models from code, structured output, testing AI answers | 🎓 | ✅ Live |
| 03 | Chat with Your Notes | Ask questions about your lecture PDFs and get answers with page numbers. Every quote is checked, and it says so when your notes don't cover a question | RAG: AI that answers from your own documents | 📚 | ✅ Live |
| 04 | German Letter Explainer | Take a photo of an official German letter. It explains what it says, what to do, every deadline with the days left and the amount to pay (not legal advice) | AI that reads images | 🌍 | ✅ Live |
| 05 | Cover Letter Agent | Plans, writes, checks and fixes cover letters and recruiter messages in English and German, and waits for your OK at every step. Nothing is ever sent |
🎓 Career & job search · 📚 Students & learning · 💼 Business & sales · 🌍 Everyone
Real screenshots with made-up sample data. Click one to open the project.
| You are… | Start with |
|---|---|
| 🎓 Looking for a job | Download the Job Application Tracker. It's free and ready to use in Excel or Google Sheets. Then check each job with the Job-Fit Checker before you apply, and let the Cover Letter Agent draft your letter |
| 📚 A student | Add your lecture PDFs to Chat with Your Notes and ask it questions. Every answer shows the page it comes from. Keep all your notes in the Second-Brain Agent to search them and connect ideas |
| 🌍 New in Germany | Take a photo of an official letter with the German Letter Explainer to see what it says, what to do and by when |
| 💼 Running a business | Let the Company Research Agent research a company and draft a personal LinkedIn note, with every fact checked |
| 🧑💻 Learning to build with AI | Open any project folder and read its README from top to bottom. Each one explains the why, not just the code |
| 🧭 A recruiter or hiring manager | See Skills this lab covers. Each skill links to the project that shows it |
| Skill | Shown in | Status |
|---|---|---|
| Python automation & testing | 01 Job Tracker: workbook generator plus 141 automated checks run in real Excel | ✅ |
| LLM APIs & structured output | 02 Job-Fit Checker: local AI with Ollama, answers forced into a JSON shape, every claim checked against the CV | ✅ |
| Testing AI answers (evals) | 02 Job-Fit Checker: 47 fast tests with a fake AI plus 11 quality checks with the real model | ✅ |
| RAG (answers from your own documents) | 03 Chat with Your Notes: embeddings with EmbeddingGemma, cosine-similarity search, answers with checked quotes and page numbers, works across English and German | ✅ |
| Multimodal AI (images → text) | 04 German Letter Explainer: the AI reads phone photos of letters (99% of words right), Python finds every date and euro amount, and the AI explains them | ✅ |
| AI agents & human-in-the-loop | 05 Cover Letter Agent: plan → write → check → fix loop with tools, two human checkpoints, and a watchdog for stuck model output | ✅ |
| Vector databases & knowledge management | 06 Second-Brain Agent: ChromaDB with incremental sync and tag filters, an agent that plans its searches, and connections between notes with checked quotes |
Every project follows the same five steps, so you always know where to look:
1. Problem → who has this problem, and why it matters
2. Build → the smallest version that actually works
3. Test → automated checks, not "it worked once"
4. Explain → README: what it does, how to run it, what I learned
5. Share → a short build log on social media
novawerk-lab/
├── README.md ← you are here
├── LICENSE ← MIT, for the code
├── LICENSE-CONTENT.md ← CC BY-NC 4.0, for the tracker, guides and images
├── assets/ ← lab banner and gallery screenshots
├── docs/ ← the lab's website (GitHub Pages)
└── projects/
├── 01-job-tracker/ ✅ live
├── 02-job-fit-checker/ ✅ live
├── 03-chat-with-notes/ ✅ live
├── 04-letter-explainer/ ✅ live
├── 05-cover-letter-agent/ ✅ live
├── 06-second-brain/ ✅ live
├── 07-company-research/ ✅ live
└── 08-first-product/ 🔜 next
Every AI project (02 to 07) uses the same layout inside its folder:
02-job-fit-checker/
├── README.md ← what it does, how to run it, what I learned
├── app.py ← the web page you click through (Streamlit)
├── jobfit.py ← the main logic: AI calls plus Python checks
├── evaluate.py ← quality checks with the real AI
├── tests/ ← fast automatic checks, with a fake AI
├── samples/ ← made-up CVs, job ads, letters or notes to try it with
├── docs/images/ ← banner and screenshots for the README
└── pyproject.toml ← the Python packages it needs (installed with uv)
Got an idea for a project, or found a bug? Open an issue. I read every one.
| Agents with a human in the loop |
| 🎓 |
| ✅ Live |
| 06 | Second-Brain Agent | Searches your notes by meaning, answers questions from them and discovers connections you haven't linked yet. Works with Obsidian | Knowledge management, vector databases | 📚 | ✅ Live |
| 07 | Company Research Agent | A team of AI agents reads a company's website, checks every fact and drafts a personal LinkedIn note or email you approve | Several agents working as a team | 💼 | ✅ Live |
| 08 | First Real Product | The most popular project, turned into a small web app that real people use | Shipping, users, feedback | 🌍 | 🔜 Next up |
| ✅ |
| Multi-agent systems | 07 Company Research Agent: Scout, Researcher, Fact-checker, Strategist and Writer with typed handoffs and Python checks between them; a polite, robots.txt-aware web crawler | ✅ |
| Deploying a real AI product | 08 First Real Product | 🔜 |