Personal Academic Knowledge Base

The Problem: Fragmented Academic Workflows

Academic researchers often juggle multiple disconnected tools: databases for discovery, note-taking apps for capture, and graph tools for synthesis. This fragmentation means search results and insights are rarely consolidated into a reusable knowledge base. Each new project risks redundant searches, and previously encountered papers or ideas become difficult to reconnect and recall effectively.

Core Capabilities & Workflow

This skill establishes a local, academic-focused personal knowledge hub centered on three core functions: capture, search, and connect.

1. Unified Data Ingestion (Capture): Consolidates knowledge from four key sources into a structured local database:

   • SmartLib Search Results: Integrates with the global literature API. After searching, you can save structured metadata (including DOI, abstract, source links) to the local db/literature.json with a single command.
   • User-Uploaded Files: Directly ingest PDFs, Word docs, Markdown, or BibTeX files. The system auto-extracts metadata.
   • News & Articles: Capture important academic news or blog posts via URL or pasted text.
   • Personal Data: Store experimental datasets, CSV files, and research notes.

2. Dual-Track Retrieval (Search): Employs two complementary search strategies:

   • Vector Semantic Search: Text (e.g., abstracts) is embedded into vectors, enabling search based on conceptual similarity, ideal for finding related papers that don't share keywords. Supports free (SiliconFlow BGE-M3) and paid high-precision models.
   • Traditional Keyword Search: Uses jieba segmentation for precise term, title, or author matching. Results from both methods are merged and presented together.

3. Automated Knowledge Graph Maintenance (Connect): Built on the LLM Wiki paradigm, the skill automatically maintains Wiki knowledge pages after ingestion. It links concepts using [[wiki-links]] and, upon batch ingestion of ≥5 documents, triggers a Lint health check to detect contradictions, outdated concepts, and orphaned pages, fostering an organized knowledge structure.

Important Boundaries & Design Choices

• Quota & Cost Model: External literature searches (via SmartLib API) consume a shared quota. However, all local operations—internal search, management, exports—are free and unlimited. When the quota is exhausted, only external searches are paused; local functionality remains fully operational.
• Data Sovereignty & Security: All data resides locally under ~/.workbuddy/academic-kb/. Raw source files in the raw/ directory are immutable. API keys for vectorization services are stored locally and sent only to the chosen provider.
• Isolated Research Sessions: A "research session" creates a read-only snapshot of literature from the main knowledge base. You can take notes and analyze within this session without affecting the primary library's data or Wiki structure, enabling focused work on distinct projects.