Systems exhibition01 / Problem 02 / Architecture 03 / How it works 04 / Failure thinking 05 / Reflection Inspect the sourceSelf Drill on GitHub
Learning system · Deterministic question engine
Self Drill
A browser-only interview practice tool that converts structured evidence into tiered questions without an API or backend.React 19TypeScriptVitepdf.jscompromiselocalStorage
resumeextractquestionreflectrepeat
The constraint.
Generate practice that responds to a person’s actual projects and trade-offs instead of serving the same static list or hallucinating through an LLM.
How it is shaped.
Client-side PDF extraction and lightweight NLP produce reviewable structured data. Five pure generation techniques build tiered questions; localStorage keeps practice history and revisit queues.
Decisions made explicit.
- Human confirmation before generation
- Rule-based techniques with maintained dictionaries
- Static fundamentals separated from evidence-driven rounds
Where it can break.
No upload leaves the browser and no network service is required. Users can correct extraction before it influences a practice session.
What I learned.
A deterministic engine can still feel personal when its rules model the right evidence and trade-offs.