arjun10g/na-tech-jobs · Datasets

arjun10g/na-tech-jobs · Datasets at Hugging Face na-tech-jobs A working ML platform for the senior data-science / ML hiring market in the US and Canada. Weekly ingest from public ATS APIs, four trained models, hybrid retrieval, and an LLM analytics layer. Runs on a $9/month Hugging Face Space. I started building this for my own job search. Each piece exists because I had a question I couldn't answer with LinkedIn or a spreadsheet: what's the actual salary distribution for senior MLE roles in Toronto, which companies are hiring at staff level right now, where do my skills line up with what's open. The ingestion cron, the salary regressor, the seniority and role-family classifiers, the matcher, the analytics tab — each one is the answer to one of those questions. 🔗 Live links | 🚀 Demo Space (dark theme, 5 tabs) | https://arjun10g-na-tech-jobs.hf.space | | --- | --- | | 📦 Source code | https://github.com/Arjun10g/na-tech-jobs | | 📊 Dataset (12,334 active jobs, weekly) | https://huggingface.co/datasets/arjun10g/na-tech-jobs | | 🧠 Models on the Hub | salary· seniority· role_family· skills | What's live on the Space Five tabs, all hitting real data and real model output. Salary. Paste a job description, the regex cascade extracts ~20 features, an XGBoost regressor predicts the maximum salary in USD/year. Edit any extracted field and the prediction updates. Held-out test-MAE is $29,091, MAPE 14.7%. Search. Substring match on title and company, with country and role-family dropdowns. It's the boring tab — useful for spot-checking the corpus before reaching for the matcher. Matcher. Natural-language query (or a pasted resume blurb) goes through dense Qdrant retrieval, an optional cross-encoder rerank, and parent-chunk hydration. You get a ranked table of jobs with a clickable apply link, plus a short LLM-generated paragraph explaining which 1-2 jobs best fit and why. Recall@10 against a labeled query set is 0.486 with rerank, 0.363 without. Analytics. Plai