SkillSphere-Agent

# SkillSphere — Hypergraph-Powered Professional Knowledge Base *A reproducible, open-source playground that turns plain career notes into a query-ready knowledge graph **and** can spit out job-targeted, ATS-friendly résumés on demand.* --- ## ✨ Why this repo exists | Pain point | How SkillSphere helps | | ----------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------- | | Career data scattered across docs, LinkedIn and slide decks | **Single-source-of-truth**: all jobs, projects and certs live in Markdown | | Recruiters can’t see proof of specific skills quickly | **Hypergraph-of-Thought** in Neo4j lets agents answer: *“Show projects proving Kubernetes cost-optimisation.”* | | Tailoring a résumé for every role is tedious | **Graph-to-PDF pipeline** converts the same graph into a fully **ATS-optimised CV** aligned to a chosen job spec | | Privacy / cost worries around SaaS LLMs | Runs **entirely local** on Ollama; no OpenAI key required | If you’re exploring AI-driven personal knowledge graphs *or* need a quick way to generate job-specific CVs, clone the repo, drop in your own records and you’ll have a live graph **and** résumé builder inside 10 minutes. --- ## 🚀 Quick start ```bash git clone https://github.com/bprager/SkillSphere.git cd SkillSphere python -m venv .venv && source .venv/bin/activate pip install -r requirements.txt # faiss-cpu, langchain-community, neo4j-driver… ollama pull gemma3:12b # or your favourite local LLM cp .env.sample .env # adjust Neo4j creds if needed # 1) Build / refresh the graph poetry run python hypergraph/ingestion_worker.py # 2) Generate an ATS-ready résumé for a target job description python scripts/build_resume.py --job-spec docs/job_postings/google_se_iii.md ``` * Browse the graph → [http://localhost:7474](http://localhost:7474) (Neo4j) * Query skills via MCP API → `curl localhost:8000/query -d '{"prompt":"List cloud skills"}'` * Résumé PDF appears in `output/` ready to send. --- ## 🗺️ Repo tour | Path | Purpose | | -------------------------------- | -------------------------------------------------------------------------------- | | `docs/` | Markdown records (`jobs`, `extras`, `certifications`) **and** job-spec examples. | | `hypergraph/ingestion_worker.py` | Ingestion pipeline with **gleaning loop** + **Node2Vec**. | | `architecture.md` | Design spec & PlantUML flow. | | `scripts/build_resume.py` | Graph→Markdown→Pandoc pipeline for **ATS PDFs**. | | `templates/` | Pandoc résumé / cover-letter templates. | --- ## 📚 Research foundation SkillSphere’s hypergraph model is inspired by: > **Haoran Luo, Haihong E, Guanting Chen, et al.** > *HyperGraphRAG: Retrieval-Augmented Generation with Hypergraph-Structured Knowledge Representation.* > arXiv: 2503.21322 (2025). [https://arxiv.org/abs/2503.21322](https://arxiv.org/abs/2503.21322) We adapt it to a **personal** graph and add: * Incremental ingest with SHA-256 change tracking. * Local-LLM **gleaning loop** that wrings \~25 % extra facts per chunk. * Neo4j GDS **Node2Vec** embeddings for structural search. * A résumé generator that queries the graph and compiles an **ATS-optimised CV** for any job description. --- ## 🤝 Why you might care * **Hiring for AI / knowledge-graph talent?** — this is a live sample of my architecture, Python and graph-data chops. * **Building internal talent graphs or CV automation?** — fork it, swap Markdown for HR data, and you’re halfway to a skills matrix and auto-CV tool. * **Just curious?** — open a PR or start a discussion; I love geeky graph & GenAI conversations. --- ## 📬 Let’s connect * Web [https://www.prager.ws](https://www.prager.ws) * Email [bernd@prager.ws](mailto:bernd@prager.ws) · LinkedIn [@berndprager](https://www.linkedin.com/in/berndprager) * Book a chat [https://calendly.com/bernd-prager/30min](https://calendly.com/bernd-prager/30min) --- © 2025 Bernd Prager — Apache 2.0 • Clone, adapt, and let me know what you build!