--- title: CareerMatch AI emoji: 🎯 colorFrom: indigo colorTo: blue sdk: static pinned: false license: mit short_description: Multilingual CV matching and skill-gap analysis tags: - nlp - multilingual - career - human-resources - data-analysis --- # CareerMatch AI | Browser Edition **Author:** Rodrigo Pantoja **Space:** `Hugdrigo/careermatch-ai` **Deployment:** Static, free web hosting, no server or paid API required. CareerMatch AI compares CV text with a vacancy description and displays an **explainable textual alignment score**, detected skills, textual evidence for requirements, shared keywords, language mentions and responsible improvement suggestions. > **Transparency:** This static browser edition uses lexical TF-IDF comparison and curated multilingual skill aliases. It **does not** load or run a transformer model, sentence embeddings, or a generative API. The original separately developed Python/Gradio version includes an optional multilingual sentence-transformer model and file parsers, but it is not what this static Space is executing. ## Try it 1. Open the **App** tab. 2. Click **Load example**. 3. Click **Analyze alignment**. 4. Review the score component breakdown, skill overlap, requirement evidence and suggestions. 5. Download a simple text report if desired. Paste text directly. Optional file upload supports only `.txt` and `.md`. For `.pdf` or `.docx`, manually copy and paste the document text. ## How scores are calculated | Component | Weight | Method | |---|---:|---| | Lexical alignment | 35% | TF-IDF cosine similarity on unigrams and bigrams | | Detected skill overlap | 30% | Multilingual curated alias taxonomy | | Requirement evidence | 20% | Heuristic shared term coverage and lexical similarity | | Keyword overlap | 15% | High-signal job-word overlap | When explicit languages are required, a small adjustment is applied for languages not mentioned in the CV. No actual proficiency is validated. When there are no detected job skills, skill coverage uses a neutral baseline, **not** an indication of competence. ## Limitations and responsible use - **Not an ATS score, hiring probability, or automated hiring decision.** No objective ground-truth label is available for a candidate. - The score measures *textual similarity/evidence only*. Strong candidates may score poorly if job and CV use different phrasing. High scores can occur for keyword-stuffed documents. - Cross-language alignment is restricted by lexical overlap; multilingual skill aliases cover only explicitly listed variations. - Missing skill terms may mean **not written in the CV**, not absent in reality. - Years of experience, licences, visas, recommendations, achievements and real competence are not independently verified. - This application runs in the browser and does not send CV contents to an application backend. Use caution with confidential documents and browser extensions. - PDF and DOCX ingestion, and true sentence embeddings, belong to the separate Python/Gradio edition and are **not available here**. ## Source layout - `index.html`: responsive UI and styling - `engine.js`: deterministic matching logic and multilingual taxonomy - `app.js`: input handling, visualization and downloadable text report - `LICENSE`: MIT licence ## Validation Engine tests should cover empty inputs, identical document similarity, absent/present multilingual skills, and score bounds. After uploading to Hugging Face, test the public app with **Load example** and **Analyze alignment**. ## Roadmap - Compare this lexical baseline to an actual multilingual embedding model on a documented benchmark. - Add more skill aliases based on real job posting datasets with proper usage rights. - Explore fairness/robustness checks and limitations across languages. - Offer additional file formats only after secure in-browser parsers are implemented. ## Related portfolio work [ResearchLens](https://huggingface.co/spaces/Hugdrigo/researchlens) is a separate multilingual document-analysis application.