Parsing in the browser
Client-side PDF/DOCX/TXT/MD parsing removes the upload entirely — the strongest privacy design is the one where the data transfer does not exist.
The Examiner's Desk · room 06 of 11
RIGOR
Curious what your resume is actually saying?
RIGOR is a browser-based technical-interview workbench: resumes are parsed entirely client-side, job descriptions are analysed deterministically with explainable outputs, and matching is evidence-backed rather than a confidence score.
It is privacy-first by design — anonymous sessions, transient resume text — and every AI-flavoured capability is optional and consent-gated.
the lamp board — requirement ↔ evidence (specimen rows)
required: 2+ yrs Python back-endevidence: built FastAPI services with 499-test suite
required: retrieval / RAG experienceevidence: Chroma + FTS5 hybrid retrieval engine
required: Kubernetes in productionevidence: no production Kubernetes deployment found
specimen rows — the workbench shows the resume lines behind every match
The actual problem
Interview preparation tools tend to hide their reasoning: a resume goes in, a verdict comes out, and the candidate cannot see why. Uploading a resume to a server for parsing also means handing your most personal document to someone else's infrastructure.
The workbench problem is to keep the entire review local and legible: parse without a server, match without a black box, and practise against a question bank that was actually reviewed rather than scraped.
The exploded machine
Resumes are parsed in the browser (PDF, DOCX, TXT and Markdown) — the document never leaves the machine, and its text is treated as transient session data.
Clip reader — in-browser resume parsing
PDF, DOCX, TXT and Markdown are parsed client-side; the document never leaves the machine.JD analysis bench — deterministic
Requirements are extracted and compared deterministically — the same inputs always produce the same analysis.Evidence-match lamp board — explainable
Every requirement match shows the resume lines behind it; a match can be challenged because its evidence is visible.Question drawer — 46 reviewed questions
Ten role families and twelve learning modules, actually reviewed rather than scraped.Session ledger — anonymous, transient
No accounts, no stored resume text — the review lives in the session and then it is gone.
Job-description analysis is deterministic and explainable: requirements are extracted and matched against parsed resume evidence with the supporting lines visible, so a match can be challenged and checked.
Preparation runs on a reviewed bank — 46 questions across 10 role families and 12 learning modules — with practice sessions in four modes: guided, technical mock, resume deep-dive and behavioural.
Engineering decisions
Explainable matches over scores
Evidence-backed matching shows the resume lines behind each requirement match; a number without its evidence is an opinion.
AI features behind consent
Optional OpenAI-backed features are consent-gated and off by default — the workbench is fully useful without them.
Anonymous, transient sessions
No accounts, no stored resume text — the review lives in the session and then it is gone.
Proof — verified results
- Question bank
- 46 reviewed questionsrepo
- Role families / modules
- 10 families · 12 learning modulesrepo
- Practice modes
- guided · technical mock · resume deep-dive · behaviouralrepo
- Resume formats parsed client-side
- PDF · DOCX · TXT · MDrepo
- Frontend
- live at rigor-flame.vercel.appverified live
every number here carries its source — my resume's figures and the synthetic-fixture results are marked for what they are