🌿 Root to Fruit — Master Project Hub

From first line of code
to career that changes a life.

Single source of truth for Yarashi's architecture, sprint plan, strategic vision, and every decision made in the build. Not just what we built — why we built it this way.

Updated 16 Jul 2026

Sprint 2 Active

Status Beta

Repo agilefoundry/agf-job-match

Session 10 hours today

🌿 Root to Fruit →

Sprints complete

Sprint 0 + 1

Active sprint

Foundation fixes

Planned sprints

Through market leadership

Bugs fixed today

Hydration, staging, deletion

Critical open

create-manual, hydration, fonts, delete

Platform

Live

yarashi.com.au

Today's session — 21 June 2026

● What we built and decided in 10 hours

LinkedIn content strategy4 posts drafted and refined: gym post, database architecture, Big Data is over, Chinese EV analogy. Posting cadence set: 24-48hr between posts.

Root to Fruit visionFull 5-layer platform architecture defined: Foundation, Intelligence Core, Journey Engine, AI Tools, Human Outcomes. Career companion vision formalised.

Intelligent data layerCanonical enrichment pipeline designed: 5-stage ingest-normalise-enrich-intelligence-deliver. Skill taxonomy, market salary tables, zero AI cost per job.

locationResolver.ts builtSingle canonical source of truth for all job API integrations. Country-first Jooble format (Country, State, City) prevents Sydney NS, Perth Scotland, Newcastle UK ambiguity.

Jooble US jobs bug fixed30 Jamestown ND jobs were entering job_feed. Root cause: pref.location was null, Jooble defaulted to US results. Fixed and deployed.

saved_jobs cleaned97 duplicate test entries removed via targeted SQL. Kept one record per unique title+company. Real saved jobs intact from 13 June onwards.

Project board restoredDocs deploy broke routing (Next.js pages in a static site). Reverted dd4da22, rebuilt as pure static HTML. Board back up within minutes.

Non-ASCII build failures fixedlocationResolver had em dashes and arrows in comments causing webpack syntax errors. Two failed builds. Stripped to pure ASCII. Lesson: always validate before shipping.

China strategy developedSea Turtle segment, government venture model, IP fork licence, personal safety exit structure, technology diplomacy framing. Full storyboard in Founder Log.

Full platform audit764KB page.jsx, 13,153 lines, 16 modules mapped. 4 critical bugs, 5 incomplete features, 6 vision gaps identified and prioritised.

Core documents

🌿

Root to Fruit — Vision & Roadmap

The complete product vision. Five-layer architecture from Foundation to Human Outcomes. Every sprint from 2 through 9. Career companion defined. The platform that walks alongside a person for life.

Vision21 Jun 2026

🧠

Intelligent Data Layer Architecture

The competitive moat. Five-stage enrichment pipeline. Enriched job_feed schema. Skill taxonomy (2,000+ skills). Market salary tables. pg_cron schedule. Zero AI cost per job. The database as intelligence, not storage.

Architecture21 Jun 2026

📖

Founder Session Log — 21 Jun 2026

The storyboard. Every strategic discussion captured verbatim: LinkedIn strategy, Big Data argument, Chinese EV analogy, China geopolitical strategy, IP fork model, personal safety decisions, guiding principles. The full 10-hour record.

Storyboard21 Jun 2026

🔍

Full Platform Audit — June 2026

Complete audit of page.jsx (764KB, 13,153 lines, 16 modules). All critical bugs, incomplete features, technical debt, and the 6 vision gaps that separate what Yarashi is from what it must become.

Audit21 Jun 2026

🚀

Beta Exit Criteria

22 criteria across 6 categories. localStorage persistence. Live progress tracking. The gate between beta and public launch. Sprint 2 must close the majority of these.

LiveActive

📋

Master Handover Document

Complete technical reference. All env vars, API keys, Supabase tables, RLS policies, cron jobs, deployment config, known issues, and safe copy baselines.

ReferenceActive

⚠

Risk Register

Active risk tracking across technical, business, security, and compliance categories. RAG status, mitigations, and owners for every identified risk.

ActiveActive

🏛

ISO Certification Pathway

ISO 27001, 9001, 42001 framework. R&D Tax Incentive qualification tracking. Dual-use: Australian compliance and Chinese government partnership credibility. Begins Sprint 8.

Sprint 8Planned

Sprint plan

SPRINT 0

Done

Security & Infrastructure Foundation

Stripe webhook verification, CRON_SECRET auth, CSP/security headers, Supabase 2FA, PII stripping across 6 AI routes, project board 2FA gate

Handover

SPRINT 1

Done

Legal Compliance & Quick Wins

Privacy Policy, ADM disclosure, LinkedIn data warrant in ToS, CV upload consent, UptimeRobot monitoring, platform_health table, extended job fields, 2-week scoring cutoff

Handover

SPRINT 2

Active

Fix the Foundation

4 critical bugs · 5 feature completions · locationResolver.ts deployed · Jooble US jobs fixed · saved_jobs deduped · project board docs added · platform audit complete

Audit Tasks

SPRINT 3

Planned

Destination Engine + Intelligent Data Layer

Career destination (user_destination table) · skill_taxonomy 2,000+ skills · enriched job_feed schema · normalisation pipeline (title, salary, location) · market intelligence tables · pg_cron schedule

Architecture Vision

SPRINT 4

Planned

Career Roadmap Module

AI-generated personalised roadmap · milestone tracking · career timeline (career_timeline table) · persistent narrative · roadmap API route · progress celebration · roadmap refresh on demand

Vision

SPRINT 5

Planned

Learning Navigator

Gap-to-course recommendations · certification pathways · free vs paid ranked by ROI · Australian training context (TAFE, VETASSESS, ACS) · skills progression tracking · learning_recommendations table

Vision

SPRINT 6

Planned

Proactive Intelligence Layer

user_alerts table · new job notifications · salary shift detection · trending skill alerts · CV staleness nudges · application follow-up reminders · career health score · My Hub daily brief · next best action

Data Layer

SPRINT 7

Planned

Career Companion AI Layer

Companion memory across all tools · context-enriched AI prompts (CV + destination + roadmap + market data) · cross-tool intelligence · career companion chat · progress-aware responses

Vision

SPRINT 8

Planned

Platform Maturity

page.jsx split into 14 module files + components + utils + constants · ISO 27001/9001/42001 documentation · R&D Tax Incentive tracking · performance audit · accessibility audit

ISO

SPRINT 9

Planned

Internationalisation & Australian Depth

Full UI translation audit across 8 languages · AI response language matching · RTL layout for Arabic · state-specific content (NSW/VIC/QLD/WA) · visa-aware guidance · TAFE/VETASSESS integration

Vision

Sprint 2 — active tasks

Critical bugs — fix before anything else

create-manual to build-cv silent failureCVTailorModule line 3392. Async chain between create-manual success and build-cv call swallowing error silently. Hydration fix is leading theory. Live test then targeted fix.

Critical

Hydration regression testcf55673 deployed but not fully regression-tested across all 16 modules. CV Maestro useEffect localStorage at line 2878 specifically needs verification.

Critical

Duplicate @import in GlobalStylesLine ~12494. Two @import statements in same style tag. Only first is honoured by browsers. Cormorant Garamond and Montserrat may be silently failing on My Hub.

High

Delete account APILine 9353 is a TODO stub showing a toast to contact support. Build /api/account/delete with 30-day grace period. Required under Australian Privacy Act.

High

Feature completions

CV Maestro update flowMake Refresh my Master CV path prominent on existing step. Clear path to re-upload LinkedIn PDF when profile changes.

Medium

CV Tailor two-step flowCurrently parallel tabs. Redesign as sequential: Feedback first, then user chooses Rewrite. Removes confusion between modes.

Medium

Pipeline UI polishCard density, status visibility, expiry date prominence, mobile layout. Kanban is functional but not polished.

Medium

Reports module decisionNav item points to MyHub. Decide: dedicated Reports component or merge permanently. Remove orphaned nav item either way.

Medium

Spanish end-to-end testTranslations exist at line 12289. Auto-detect and manual selector untested. Many modules still use hardcoded English strings instead of t. references.

Medium

Completed today in Sprint 2

locationResolver.ts deployedCanonical location library built. Country-first Jooble format (Country, State, City) across 57 locations. Prevents Sydney NS/FL, Perth Scotland, Newcastle UK, Adelaide SA ambiguity. buildAdzunaUrl, buildJoobleBody, isLocationConsistent exported. All API routes now import from single source.

Done

Jooble US jobs root cause fixed30 Jamestown ND jobs in job_feed. Root cause: pref.location was null, passed empty string to Jooble, returned US results. Fixed with explicit city resolver. Non-ASCII build failures (em dashes in comments) and orphaned closing brackets also fixed across 3 deploys.

Done

saved_jobs database cleaned100+ duplicate test entries removed. incrementinc role appearing 30+ times from create-manual silent failure bug testing. Test entries (ServiceNow PM, SN PM, Test, 1 IT Project Manager) removed. 97 clean real saved jobs retained.

Done

Project board routing fixedDocs deploy commit dd4da22 added Next.js app/ folders to a static HTML site, breaking Vercel routing and causing 404 on entire board. Reverted immediately. Docs rebuilt as pure static HTML files matching existing site structure.

Done

Strategic sessions — captured today

🌎

China Strategy — Technology Diplomacy

Yarashi identified as having no equivalent in China from the candidate perspective. All Chinese platforms (Boss Zhipin, Zhaopin, Liepin) serve employers, not individuals.

Primary segment: Sea Turtles (Chinese nationals returning from Australia with international CVs)
Entry model: IP fork licence — keep Yarashi brand and codebase in West, sell China a rebranded copy with geographic exclusivity
Community model: 70% of profits to boxing clubs and youth retraining
Safety structure: No executive role, no board position, no data access after transfer — eliminates espionage risk to family
Precedent value: First Western SaaS founder to cleanly sell to Chinese government sets terms for future transfers
Infrastructure: Alibaba Cloud + self-hosted Supabase + Qwen/ERNIE AI + Alipay + WeChat Pay

🚀

LinkedIn Thought Leadership — 4 Posts

Four posts developed, refined, and pressure-tested for accuracy today. Posting cadence: 24-48 hours apart.

Post 1 (Personal): Claude told me to go to the gym — Yarashi built with PM discipline: sprints, burndown, beta exit strategy
Post 2 (Technical): Legacy databases were never designed for AI — bolted-on vs AI-native architecture
Post 3 (Bold): The era of Big Data is over — it was a sticking plaster on a structural wound
Post 4 (Analogy): Chinese EV analogy — BYD clean-sheet vs European combustion platform. Applied to Oracle/Salesforce/ServiceNow. Closing word: "Catching up."

All Yarashi architectural details deliberately omitted for competitive reasons.

🧠

Intelligent Data Layer — The Real Moat

Every job board worldwide produces identical raw fields: title, company, description, location, salary, date, url. The moat is what happens to that data before the user sees it.

Stage 1 Ingest: Any source, source-agnostic normalisation contract
Stage 2 Normalise: Title lookup table, salary parser, location geocoder — zero AI
Stage 3 Enrich: Skill taxonomy extraction, seniority detection, industry classification — zero AI
Stage 4 Intelligence: trending_skills, market_salary, role_demand, company_intelligence — pg_cron, zero AI
Stage 5 Deliver: Pre-computed, sub-100ms, proactive — user never waits

AI fires only for narrative generation. Output stored permanently. Never repeated for same user + context.

🗺

Career Companion Vision — Root to Fruit

Kevin's founding statement: "It must NOT be static. It must take people into a career, on a career journey and recommend them valuable things like training programmes, courses, skills which will make them stand out, put them in the right direction from school to career."

Destination Engine: Where do you want to be in 12 months? Sets north star for everything
Career Roadmap: AI-generated personalised path from today to destination. Persists and evolves
Learning Navigator: Gap identified + course recommended, ranked by ROI, Australian context
Career Timeline: Every action, application, skill, milestone recorded chronologically
Proactive Intelligence: Platform pushes insights before user asks — never static

Not a job board. Not a CV tool. A career companion for life.

System architecture

Five-layer career intelligence system Full vision →

🌿

Foundation

Auth, DB, RLS, Security, ISO, Billing, Monitoring

🧠

Intelligence core

Destination engine, Market intel, Profile intel

🗺

Journey engine

Roadmap, Learning Navigator, Timeline, Alerts

⚡

AI tools

CV Maestro, CV Tailor, Skill Radar, Interview, LinkedIn

🍋

Human outcomes

Migrants placed, Careers built, Communities stronger

Data enrichment pipeline — zero AI cost per job Full architecture →

📥

Ingest

Adzuna, Jooble, Seek, Indeed — same raw fields everywhere

🔧

Normalise

Title, salary, location via lookup tables. Pure logic, no AI.

✨

Enrich

Skills extracted, seniority detected, industry classified

💡

Intelligence

CV match (SQL), alerts (pg_cron), health score — zero AI

🚀

Deliver

Pre-computed, instant, proactive. User never waits.

Location resolver — canonical source of truth Technical ref →

📍

resolveLocation()

Converts any user input to fully qualified config

🏭

buildAdzunaUrl()

Correct country endpoint + where param

🌐

buildJoobleBody()

Country, State, City format — prevents Sydney FL, Perth Scotland

✅

isLocationConsistent()

Post-fetch filter catches anything that slipped through

📄

57 locations

AU, UK, IE, US, CA, NZ, SG, UAE — all fully qualified

New database tables — Sprint 3 onwards

skill_taxonomy

2,000+ skills with aliases, categories, related skills, industry relevance, seniority signals, and rolling 7/30/90 day demand counts. The dictionary every platform feature reads from. Built once, grows forever.

Sprint 3 · Core

trending_skills

Skill demand aggregated from job_feed over 7/30/90 day windows. Trend direction (rising/stable/declining), average salary for roles requiring this skill, top industries and locations. Refreshed every 4 hours by pg_cron. Zero AI.

Sprint 3 · pg_cron

market_salary

P25/P50/P75/P90 salary bands by role family, seniority level, location region, and industry. Computed directly from real job_feed data. No surveys, no third-party data. Recalculated every 4 hours.

Sprint 3 · pg_cron

user_destination

Target role, industry, salary range, timeline (6/12/36 months), and career stage (school-leaver, graduate, early, mid, senior, career-changer, returner). The north star that drives all personalisation.

Sprint 3 · New

career_roadmap

AI-generated personalised milestone path from today to destination. Stored and versioned, not regenerated every visit. User marks milestones complete. Platform celebrates progress.

Sprint 4 · New

career_timeline

Chronological record of every significant career event: CV uploaded, job applied, skill added, course completed, interview prepped, milestone hit. The persistent narrative of each user's journey.

Sprint 4 · New

user_alerts

Proactive notifications: new matched jobs, salary shifts, trending skills missing from CV, CV staleness after 60 days, application follow-up nudges after 14 days. Generated by pg_cron. Zero AI cost.

Sprint 6 · pg_cron

career_health

Composite 0-100 score: CV freshness (0-25), market alignment (0-25), application activity (0-25), skill growth (0-25). Pure SQL calculation from existing data. Recalculated daily.

Sprint 6 · pg_cron

Guiding principles — every decision runs through these

Serve the person, not the metricEvery feature must genuinely help the user. If it doesn't serve the person, it doesn't ship.

Proactive over reactiveYarashi tells you what you need before you ask. The platform works while the user sleeps.

Honest over flatteringTell users where they actually stand. Soft lies waste their time. Yarashi is a trusted advisor, not a cheerleader.

Persistent over transactionalEvery interaction builds the journey. Nothing is a one-off. The platform accumulates meaning.

Simple over impressiveIf a user needs to read instructions, the feature has failed.

Inclusive by defaultSchool-leavers, migrants, career-changers, returners — all first-class citizens, always.

Built to lastISO certified. Documented. Auditable. Worthy of the trust people place in it.

The database is intelligenceNot storage. Every row enriched before the user sees it. The data layer is the product.

AI used sparingly, stored permanentlyAI fires only where lookup tables cannot. Every output cached. Never repeated for same context.

Generosity as strategy70% back to the people — boxing clubs, retraining, community. Whether in Sydney, Ireland, Spain, or China.

"Our purpose is to serve the people. Help people. From root to fruit." — Kevin Delaney, Founder, Yarashi · 21 June 2026

Document cross-references

From

Connects to

Document

Vision & Roadmap

Sprint 2-9 task breakdown

vision

Platform Audit

Sprint 2 critical bugs and priorities

audit

Data Architecture

Sprint 3 database build tasks

data-architecture

Data Architecture

Sprint 6 proactive alert system

data-architecture

Founder Session Log

All strategic decisions and analogies

founder-log

Beta Exit Criteria

Sprint 2 completion gate — 22 criteria

beta-exit

Risk Register

All sprints — risk tracking

risk-register

Handover Document

Technical reference for all sprints

handover

ISO Pathway

Sprint 8 certification documentation

iso

locationResolver.ts

lib/locationResolver.ts in agf-job-match repo

handover

From first line of codeto career that changes a life.

From first line of code
to career that changes a life.