Sospana Sonke
Sospana Sonke stores jobs, internships, learnerships, bursaries and graduate programmes from official pages, lists employers with a direct careers link, shows the Statistics South Africa labour table, and answers a search only from what is stored. You apply on the source page. The tools are free. We do not promise a place.
The public line is: Discover what's possible. Understand what's changing. Act on what matters.
The sections below describe the original Phase 1 foundation. Several of those claims are no longer how the product works. Job matching was removed. The employer directory is far larger than the 367-company seed named here. There is no paid gate on the public tools. See CLAUDE.md for the system that is running.
This repository contains the Phase 1 foundation of the platform described in the
Product & Technical Blueprint. It is a real, running, tested codebase — not a mockup.
What works today (Phase 1 foundation)
- Authentication — register, email verification, login, token refresh,
/me, with Argon2 password hashing and JWT tokens.
- Role-based access control — candidate vs administrator.
- Company database — the JSE + State-Owned-Entity seed data (367 companies), with CSV import (admin-only) and de-duplication.
- Careers-URL tester — validates that a stored careers URL resolves and looks like a careers page, and updates each company's status (this is how all links get verified at scale).
- Candidate profile — full profile plus education, certifications, work experience, and skills, each ownership-scoped, with a
confirmed_by_candidate flag separating verified fact from AI suggestion.
- CV upload & intelligence — upload PDF/DOCX/TXT (size, type, magic-byte and malware-scan checks), the original is stored immutably, text is extracted, and an AI provider produces a structured suggestion you can import into your profile as unconfirmed data.
- AI provider abstraction — swappable providers (a free offline heuristic by default, a Claude implementation when configured); it never fabricates candidate facts.
- Vacancy discovery engine — a strategy framework that reads vacancies from Greenhouse, Lever, and SmartRecruiters JSON feeds and from schema.org JobPosting data on static pages; respects robots.txt with rate limiting and backoff; normalises and de-duplicates vacancies by content hash; classifies each requirement as hard vs soft; detects closed roles; and exposes admin scan + vacancy-browsing endpoints.
- Matching engine — (Job matching was removed in Oct 2026: no per-user matching, match scores, strong-match alerts or match pages. The scorer is kept only to analyse one job advert a candidate pastes into the CV builder;
candidate_matches rows are untouched.) A cheap deterministic pre-filter followed by a configurable weighted scorer across eight dimensions, a hard-requirement gate, and an explainable APPLY / REVIEW / DO-NOT-APPLY decision with plain-language reasons and gaps, a confidence level, and a match band. Scoring is deterministic (never LLM-controlled); weights and thresholds are admin-editable.
- Document generation — from a match, generates a truthful tailored CV (PDF + DOCX) built only from confirmed profile data, with vacancy-relevant skills/experience floated to the top, plus a cover letter. Every document passes a truthfulness validator that rejects any skill, employer, qualification, or inflated experience figure not backed by the profile, and each CV gets an ATS compatibility score with a breakdown.
- Applications & tracking — prepares an application from a match through the full status lifecycle with an immutable audit trail; three modes (automatic / approval / assisted); anti-spam caps (per-day, per-week, minimum score, duplicate prevention, do-not-apply gate); and never-fabricated answers — factual questions come from the profile, and anything absent is flagged
UNKNOWN — CANDIDATE INPUT REQUIRED for the candidate to fill. Candidate preferences (mode, caps, exclusions) also feed the matcher.
- Subscription & payments — the R100/month subscription behind a swappable payment provider (Paystack for production with HMAC-SHA512 webhook verification; a Mock provider for offline dev/tests). A subscription state machine (TRIAL → ACTIVE → PAST_DUE → CANCELLED/EXPIRED) with a grace window, access gating that returns 402 on matching / document / application endpoints when inactive, checkout, cancel, and idempotent webhook handling.
- Dashboards & reports — a candidate dashboard (vacancies found, strong matches, CVs, applications, awaiting-action, interviews, offers, subscription) and an admin/business dashboard (registered candidates, active + paying subscriptions, estimated MRR, company/source health, vacancy and application counts), plus a per-cycle candidate intelligence report rendered to a downloadable PDF.
- Scheduler & notifications — a swappable email provider (console default, SMTP for production), idempotent dashboard notifications (and optional email) for strong matches, action-required applications, and ready reports, plus recurring jobs (scan-all-companies, match-all-candidates) with a run log and an admin-configurable cron schedule (run via
python -m scripts.run_scheduled_job <job>).
- Automated tests — 112 tests covering auth, access control, CSV import/dedup, URL tester, profile CRUD + ownership isolation, the CV pipeline, the scraper, the matching engine, document generation, the application lifecycle, subscriptions, dashboards/reports, and notifications/scheduler.
Phase 1 (MVP) is complete — all 16 MVP modules plus the Next.js frontend, built, tested, and verified end-to-end.
以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗
