From first job search
to career that changes a life.

Onboarding destination flow

• New onboarding step: "Where do you want to be?" -- after profile setup. Role, industry, timeline (6/12/36 months). North star for everything.
• user_destination table -- target_role, target_industry, target_salary, target_timeline_months, current_role, career_stage
• Destination visible on My Hub -- persistent banner: "Your destination: Senior PM -- $140k -- 12 months"
• Destination update flow -- user can update at any time from My Hub or Profile

Intelligent data layer -- the competitive moat

• skill_taxonomy table -- 2,000+ skills, aliases, categories, related skills, industry relevance, seniority signals, demand counts (7/30/90d), trend direction
• Normalisation pipeline -- title_normalisations lookup, salary parser (all formats to annual integer), location_normalisations with lat/lng/region/timezone. Zero AI cost.
• Market intelligence tables -- trending_skills (P25/P50/P75/P90 by role/location), market_salary, role_demand. All computed by pg_cron every 4 hours from job_feed. Zero AI.
• Career stage detection -- classify user from CV and profile: school-leaver / graduate / early / mid / senior / career-changer / returner

Career Roadmap module

• New nav item: Career Roadmap -- AI-generated personalised path from today to destination
• Roadmap generates -- honest current assessment, the gap (skills/experience/credentials/network), 3-5 milestone steps, timeline per milestone, what success looks like at each stage
• Roadmap stored and versioned in Supabase -- not regenerated every visit. Persists. User sees their roadmap every time.
• Roadmap progress tracking -- user marks milestones complete. Platform celebrates progress.
• Roadmap refresh -- user requests new roadmap when situation changes (new job, new skill, changed destination)
• /api/career/roadmap -- takes profile + destination + CV, returns structured milestone JSON

Career Timeline

• Timeline view -- chronological record of every significant event: CV uploaded, job applied, interview completed, skill added, roadmap milestone hit
• career_timeline table -- event_type, event_data, created_at. Auto-populated by every tool action.
• Timeline on My Hub -- "Your career journey so far" -- shows progress, builds emotional connection

• Learning Navigator -- extends Skill Radar. For each gap: free options (TAFE, Coursera), paid options (Udemy, LinkedIn Learning), fastest path, highest ROI.
• Australian context -- TAFE NSW/VIC/QLD, government funded training, industry body certifications (ACS, Engineers Australia, VETASSESS)
• Certification pathway -- for PM: CAPM to PMP to AgilePM. For cloud: AWS/Azure/GCP tracks. For cyber: CISSP/CompTIA pathway.
• /api/career/learning -- takes skill gaps + career stage + target role, returns ranked recommendations
• Skills progression tracking -- history of skills added, market alignment score, "You added 3 skills this month" on My Hub

Proactive alerts

• New roles alert -- "3 new roles matching your profile appeared today"
• Market shift alert -- "Demand for your target role increased 12% this month in Sydney"
• Salary movement alert -- "Average salary for Senior PM in Sydney moved to $145k"
• Skill trending alert -- "ServiceNow is appearing in 40% more PM roles -- it is not in your CV"
• CV staleness alert -- "Your CV has not been updated in 60 days -- the market has shifted"
• Application follow-up nudge -- "You applied to [company] 14 days ago. No response yet. Want to follow up?"

My Hub upgrade

• Hub becomes proactive, not passive -- leads with alerts and recommended actions, not static stats
• Daily brief -- "Good morning Kevin. Here is what changed in your market since you last visited."
• Next best action -- single most important thing to do today, based on roadmap and market data
• Career health score -- composite 0-100: CV freshness + market alignment + application activity + skill growth

• Companion memory -- every AI call enriched with career stage, destination, roadmap position, recent activity. Responses become deeply personalised.
• Cross-tool intelligence -- Skill Radar gaps feed CV Tailor. Interview Prep knows your weaknesses from previous sessions. Cover Letter knows your strongest selling points from CV Match.
• Career companion chat -- conversational AI knowing everything about your journey. Answers: "What should I focus on this week?", "Am I ready to apply for this role?", "How do I close this gap?"
• Not a support bot -- strategic, personal, honest career intelligence conversation

page.jsx component split

• Platform.jsx -- shell, auth, nav, routing only
• 14 module files -- Dashboard, CVMaestro, CVTailor, JobFeed, JobSearch, JobAnalyser, InterviewPrep, CoverLetter, SkillRadar, LinkedIn, Pipeline, Profile, CareerRoadmap, LearningNavigator
• components/ -- Toast, Modal, LoadingScreen, InfoBtn, shared UI
• utils/ -- fetch, cache, rate limits, exports
• constants/ -- colors, plans, nav items, help content

ISO certification documentation

• ISO 27001 -- data flow maps, risk register, access control policy, incident response plan
• ISO 9001 -- quality management process documentation
• ISO 42001 -- ADM disclosure, bias monitoring, human oversight framework
• R&D Tax Incentive -- document all qualifying activities from Sprint 0 onwards

• Full UI translation audit -- all hardcoded English strings replaced with t. references across every module
• AI response language matching -- outputs generated in user selected language
• RTL layout -- Arabic (ar) requires full RTL CSS audit
• All 8 languages end-to-end tested
• State-specific content -- NSW/VIC/QLD/WA job market data and salary benchmarks
• Visa-aware content -- skilled migrant context, work rights, credential recognition guidance
• Australian certification pathways -- TAFE, VETASSESS, ACS, Engineers Australia integrated into Learning Navigator

The idea

• Dedicated @yarashi.com.au email address per user -- issued on sign-up or opt-in. Used exclusively for job applications and recruiter communication. Keeps personal email clean and separate.
• Inbound mail stored inside Yarashi -- every recruiter reply, every rejection, every interview invite lands inside the platform. Matched to the application it belongs to. No more hunting through a personal inbox.
• Auto-correlation with ApplicationTracker -- when a reply arrives from a company the user applied to, it is linked to that pipeline card automatically. Status updated from "Applied" to "Response received".
• Compose and reply inside Yarashi -- send follow-ups, accept interview invites, and decline offers without leaving the platform. Full thread view per application.
• AI-assisted reply drafts -- "Draft a follow-up for this application" generates a context-aware reply using the original CV, the job, and the recruiter's message.
• Smart notifications -- "You received a response from Atlassian. Likely an interview invite. Open it?" Proactive nudge rather than passive inbox.

Why it matters

• Owns the most complete picture of a job search -- no other platform touches the email layer. LinkedIn has InMail, Seek has nothing. Yarashi owning the inbox means Yarashi has the full signal: applied, responded, interviewed, offered, accepted.
• Eliminates the biggest friction point -- recruiter emails get lost in personal inboxes. Users miss callbacks. A dedicated address with smart routing eliminates this entirely.
• Natural differentiation -- a @yarashi.com.au address signals professional organisation to recruiters and creates a visible anchor for the brand in every application sent.

Infrastructure required

• Inbound mail routing -- MX records on yarashi.com.au pointing to a mail service (Cloudflare Email Routing, Mailgun Inbound, or Postmark Inbound). Webhooks fire on delivery to a Yarashi API route.
• user_inbox table in Supabase -- stores from, subject, body_text, body_html, received_at, application_id (FK to saved_jobs), thread_id, read, replied_at per user.
• Sending layer -- replies sent via Resend or Postmark with the @yarashi.com.au address in From. DKIM and SPF records on yarashi.com.au required.
• Spam and deliverability -- inbox warming, reputation monitoring, bounce handling, unsubscribe compliance. Critical for @yarashi.com.au to be trusted by recruiter spam filters.
• Storage and retention policy -- email body content is personal data under the Privacy Act. Retention must be documented, consent must be explicit at opt-in, and deletion must cascade when account is deleted.

Compliance considerations

• Australian Spam Act 2003 -- storing email content triggers specific obligations. Privacy Policy must be updated to cover inbox storage explicitly with user opt-in consent at address issuance, not at platform sign-up.
• Third-party email content -- recruiter replies stored in Yarashi contain the recruiter's personal data as well as the user's. Must be scoped narrowly to the job search context and excluded from any aggregation or intelligence features without explicit secondary consent.
• Account deletion cascade -- all inbox content must be hard-deleted within 30 days of account deletion request. Extend the existing /api/account/delete cascade to cover user_inbox rows.

Near-term alternative (build first, Sprint 3 or 4)

• Application logging without email infrastructure -- when a user applies for a job from within Yarashi, auto-log it to ApplicationTracker with date, role, company, and a notes field. Captures 80% of the tracker value at zero infrastructure cost. Validates that users actually use the pipeline before committing to an email layer.

Prerequisites before building

• Platform is stable and split -- page.jsx component split (Sprint 2) and Next.js upgrade complete. No major infrastructure debt.
• Paying users exist -- validate that ApplicationTracker is actively used before adding the email layer on top of it.
• @yarashi.com.au brand is trusted -- volume and tenure needed before the address is recognised by recruiters as professional rather than suspicious.
• Privacy Policy updated -- specific consent language for inbox storage reviewed and approved before a single email is stored.

Team composition

• Maya — Researcher. User interviews, behaviour analysis, persona development. First move: 10 user conversations in 30 days. Feeds findings to Jordan and Sam before any new copy or strategy is written.
• Jordan — Strategist. Positioning, ICP, market framing. First move: one-page ICP for the Australian market. Everything the team builds is anchored to this document.
• Sam — Copywriter. Voice, landing page, onboarding, empty states. The product should feel like a smart friend, not a SaaS tool. Rewrites landing page once Jordan delivers positioning.
• Alex — Builder. Feature specifications, technical feasibility, cost-aware architecture. First move: instrument every AI call and user action — without data the team is blind.
• Riley — Marketer. Acquisition, channel strategy, 100 real beta users in 30 days. Quality over volume — users who give real feedback, not just sign-up numbers.
• Finn — Budget and cost analyst. Challenges every AI call decision. Veto rights on any feature that adds unbounded AI spend. No feature ships without a cost model: estimated calls per user per day, cost per MAU, break-even against plan revenue.
• Cleo — Creative director. Product lifecycle, design vision, long-term roadmap. Thinks in years not sprints. Reviews every sprint for design coherence. Owns the "where are we heading" brief every six weeks.

First meeting — key decisions (23 Jun 2026)

• The hook is the match score, not the platform. Jordan identified it, Sam sharpened the language, Alex said he could build it in onboarding, Riley saw it as the acquisition mechanic. "Know your match score before you apply" is cleaner and more honest than "AI-powered job search."
• Instrument before building anything new. Alex week-one priority. No event tracking means no data on which tools users actually use or where they drop off. Unblocks every other team member simultaneously.
• Free match score as acquisition hook. Anonymous users get a fast client-side approximate score — zero API cost. Authenticated users get the real Claude score. Sign-up becomes the gate to accuracy not just features. Riley: this is the acquisition mechanic. Maya: the jobs people paste in become the first research signal.
• Sign-up is the cost gate, not a friction point. Finn's framework: anonymous = zero AI calls, free authenticated = rationed calls, paid = full product. This is not a restriction — it is the business model.
• Cache key normalisation is an immediate fix. Finn identified that minor variations in how users paste a job title (punctuation, spacing) create cache misses and double API calls. Normalising the key could improve cache hit rate from 20% to 60% and cut AI spend by 40-60%.
• The moat is data that compounds, not AI features. Cleo: everyone will have AI. Peer benchmarking, company intelligence, market signals — these require the application tracking layer that Seek and LinkedIn never built. Yarashi has it. Every user action makes the product smarter for everyone.
• The page.jsx split is the foundation for the entire roadmap. Alex: the autonomous agent, career companion AI, mobile redesign — none of it can be built cleanly on a 13,000-line monolithic file. The split is not maintenance. It is the prerequisite for everything on Cleo's roadmap.
• Three-phase roadmap agreed. Phase 1 (now-3 months): trust — the product works perfectly. Phase 2 (3-6 months): intelligence — the product works for the user. Phase 3 (6-12 months): moat — the product is impossible to leave.

Cost framework — Finn's model (non-negotiable)

• Anonymous visitor: zero AI calls. Can see product, sample outputs, approximate match score only. Cost: $0.
• Free authenticated user: rationed AI calls. CV scored once on upload (cached 7 days). Job match: 5 per day. No CV Tailor, no Interview Prep, no Skills Gap. Cost: ~$0.04/MAU.
• Starter plan: CV Tailor 3/month, Interview Prep 5/month, Skills Gap 2/month. All results cached — repeat calls return cached result at $0. Cost: ~$1.20/MAU.
• Pro plan: unlimited job matching, CV Tailor 10/month, all tools. Cache-first on every repeat call. Cost: ~$3.50/MAU before caching, ~$1.40/MAU after 60% cache hit rate. At $29/month pricing: strong margin.
• Break-even: approximately 8-10 paid Pro users covers current infrastructure (Vercel Pro + Supabase + AI costs). Achievable in first month of proper beta.
• Cache-first rule (all tiers): same CV + same job = return cached result. No exceptions. Cache TTL: 7 days for scoring, 24 hours for job feeds. Normalised cache key (lowercase, strip punctuation, trim whitespace). Estimate: 40-60% reduction in AI spend at scale.

Competitive landscape — Cleo's audit

• What Yarashi already does that no competitor does: AI match scoring per role, CV tailoring per application, role-specific interview prep, full application pipeline tracking, detailed skills gap analysis.
• What Yarashi is building that no competitor has: career health score, Yarashi Inbox (@yarashi.com.au), market intelligence feed, peer benchmarking, company intelligence (verified by behaviour), career companion AI, autonomous application agent.
• The gap: Seek has a job board and basic tracking. LinkedIn has network and partial skills data. Resume.io has templates. None of them own the full stack: find job → score match → tailor CV → prepare for interview → track application → receive response → measure progress. Yarashi is the only product attempting to own the entire job search lifecycle.
• Positioning statement (Jordan + Sam): "The only job search platform that is genuinely on your side." Not the employer's side, not the recruiter's side. The job seeker's side, across their entire career.

AI department brief — Cleo

The product has to work perfectly before it works magically. Sprint 11 adds the intelligence layer that makes Yarashi feel alive — not just reactive to what the user does, but proactively working for them.

• Career health score (s11-1). Single 0-100 metric on the dashboard — CV freshness + market alignment + application activity + skill gaps. Updates weekly. User always knows where they stand. Low AI cost, high retention impact.
• Application timeline (s11-2). Visual timeline auto-populated from ApplicationTracker. Applied → responded → interviewed → offered → accepted. No extra AI calls — pure data visualisation. First product in the market that makes the job search journey visible.
• Emotional design layer (s11-3). Celebrations, coaching moments, encouragement. Product feels like a smart calm friend. Acknowledges rejection, celebrates activity, tells users they are ready. No AI cost — pure UX and copy. Sam owns the voice.
• Mobile-first redesign (s11-4). Job feed, match score, tracker redesigned for mobile. Prerequisite: page.jsx split complete. Cannot redesign a 13,000-line monolithic component.
• Market intelligence feed (s11-5). Weekly personalised digest: salary shifts, skills trending in job ads, companies hiring. Powered by Adzuna/Jooble data + Claude. Cache per cohort not per user — one compute serves thousands.
• Skill certification pathways (s11-6). Skills Gap recommends the exact certification to close each gap — name, provider, cost, time, link. Closes the loop between insight and action. High retention feature.
• Event instrumentation (s11-9). Track every AI call, cache hit, and user action. Alex week-one priority. Unblocks the entire team.
• Cache key normalisation (s11-8). Fix double-billing on repeat calls. Finn: could improve cache hit rate from 20% to 60%. 40-60% AI spend reduction.
• Free match score hook (s11-10). Approximate score for anonymous users at zero API cost. Real score behind sign-up. Strongest acquisition mechanic identified by the team.

AI department brief — Cleo

The moat is not AI. Everyone will have AI. The moat is data that compounds with every user — peer benchmarking, company intelligence, market signals that Seek and LinkedIn cannot build because they never had the application tracking layer to generate it.

• Peer benchmarking (s12-1). "Your match score is 74%. Yarashi users with your background average 68%." Aggregated, anonymised. Network value grows with every user. Prerequisite: 500+ users in same role/location cohort. Start with senior IT project managers in Sydney.
• Company intelligence layer (s12-2). Anonymous application and interview data reveals which companies actually respond, how long they take, interview stages, offer rates. Glassdoor but verified by behaviour not self-reported. Riley: "Yarashi users know which companies actually respond" — creates FOMO for non-users.
• Career companion AI (s12-3). Conversational AI with full user context — every role applied, every rejection, every interview prep, every skill gap. "What should I focus on this week?" gets a personalised data-driven answer. Prerequisite: page.jsx split + Next.js upgrade + companion memory layer.
• Sign-up as cost gate (s12-4). Enforce user ID on every AI route. 401 if null. Enables rate limiting, plan tier enforcement, cache by user, cost tracking per user. Alex: ~2 hours to enforce consistently.
• Cost dashboard (s12-5). Real-time AI spend per user, tool, and plan tier. Cache hit rate. Break-even tracker. Finn: "right now I am estimating blind. With this data I can tell Kevin exactly which features are profitable."

AI department brief — Cleo

This is where Yarashi becomes a category of one. Not an AI job search tool — an autonomous career partner that works while you sleep. None of this exists in the market. All of it is buildable on the foundation we are creating now.

• Autonomous application agent (s13-1). Yarashi finds matching roles, drafts cover letters, queues applications for one-click send. User reviews and approves. The first truly agentic job search product. Prerequisite: stable split codebase and Next.js 16. The page.jsx split is the foundation for this feature.
• Visa and migration pathways (s13-2). Skill migrant guidance, work rights, credential recognition. Underserved segment in the Australian market — completely ignored by Seek and LinkedIn. Initial cohort: skilled migrants in IT and engineering roles in Sydney.
• Employer-side product (s13-3). Once user volume exists, offer employers direct access to AI-matched candidates. Yarashi users are pre-qualified by match score. Reverses the job board model. Finn: this is the revenue model that makes the free tier sustainable at scale.

The roadmap in one sentence

Build trust first, add intelligence second, create the moat third. Every feature on this roadmap answers one question: does this make the user more confident, more prepared, and more likely to land the role they actually want?