Scope · 2026-05-08

Three projects → sellable product.

Current state, gap to ship, AI-augmented dev hours, costs, and business structure — for the audit tool, the NLP app, and Tender Hunter.

Illumen Group Pty Ltd ACN 696 920 895 One dev + AI agents in parallel Wall-clock hours, not solo-dev hours

Project 01 — Lead-gen weapon

Illumen Audit → cold-call ready

Run on any Perth SMB → walk in tomorrow with a credible, persuasive report.

Status

75%

Hours

24–32h

Nights

Cost

<$100

What works today

/opt/prospector/ — full Python pipeline. Scraper (Google Places → 546 Perth SMBs across 10 industries) → auditor (Playwright + BeautifulSoup, 480 of 487 sites audited cleanly) → scorer (rule-based buying-signal weights) → reporter (WeasyPrint 6-page PDF). 30 reports already generated for the Apr 20 walk-ins. Restartable, rate-limited, modular.

Gaps to close

No single-business CLI. Currently runs the whole 546-business batch. Need audit.py "Aria Dental Perth" → /tmp/aria.pdf.
Live verification missing. Phone, address, hours pulled from stale Google Places — credibility risk if owner says "wrong number."
Report too long & generic. 6 pages, no priority ranking, no ROI calc, no suburb percentile. Only 30/546 prospects analysed via Claude — fallback estimates feel boilerplate.
Booking-widget detection has false negatives. 20 hardcoded patterns, no JS-trigger — misses Mindbody, HubSpot, lazy-loaded widgets → wrong cold-call angle.
No SMB-pipe integration. Report sits as PDF, doesn't feed CRM or trigger follow-up.

Effort breakdown

Single-business CLI wrapper — 3–4h
Live verification module (phone/whois/DNS) — 6–8h
Competitor filtering by suburb — 2–3h
Report redesign (2-page cold-walk format) — 5–6h
SMB-pipe webhook integration — 4–5h
Smoke test on 10 real businesses — 8–10h

Cost

Almost zero new spend. Google Places API stays in cheap tier (~$10/mo). Claude analyser at scale ~$10/run. Already covered.

Why this is #1

Fastest ROI on the board. 2 nights of work → cold-walk 20 businesses → close 2–3 at $2–3k each → funds everything else. Lowest risk, fastest cash.

Project 02 — Consumer app

NLP Mastery → first-class launch

iOS + Android + web. Polished. No post-launch rework. Users genuinely enjoy it.

Status

60%

Hours

350h

Nights

15–20d

First-yr cost

~$2k

What works today

PWA architecture — vanilla JS SPA + FastAPI + PostgreSQL on port 8103. All 62 L1–L3 techniques documented in CURRICULUM.md. Onboarding flow with disclaimer + 6-track skill picker working. Subway-map skill tree, dark-on-translucent design system done. Stripe backend wired (regional pricing). 7 voice source recordings captured for ElevenLabs clone training.

Gaps to close

No native packaging. Currently PWA only. Capacitor wrap → Xcode + signing + TestFlight, Android Gradle + Play Store. ~50h.
Payment UI doesn't exist on frontend. Backend has Stripe, app has no checkout. Plus IAP for App Store compliance.
Voice coaching scaffolded but not wired. ElevenLabs Conversational AI bridge exists, coaching engine not connected to practice flow. This is the bet-the-app feature — without it, $199/yr pricing isn't justified.
Voice clone audio not generated. 30 min source recorded, but no clone trained, no per-technique guided voiceovers (62 × ~2 min).
Spaced repetition is skeleton only. Intervals defined, but no scheduler, no review queue, no L2L 5-day-streak unlock logic.
No push notifications. Required for retention loops + L2L bonus trigger.
App Store assets missing. Icon variants, screenshots, descriptions, privacy policy, ToS.
No analytics or crash reporting. Sentry, funnel tracking, session telemetry absent.

Effort breakdown

Component	Hours
Capacitor + iOS/Android packaging + store assets	70
Payment UI (Stripe + IAP hybrid)	40
Voice coaching wiring + clone + per-technique audio	100
Spaced repetition + L2L unlock	25
Push notifications	18
Analytics + crash reporting	25
Legal / WCAG / compliance	15
End-to-end QA across devices	50
Canonical essay + affiliate setup	25
Total	368

Strong middle path

Web-only first launch (skip native, ship as PWA only): ~150 hours / ~6 nights. Validate voice coaching converts paying users before burning 200 more hours and $2k on store fees. Recommended.

Cost — one-time

Apple Developer Program: US$99/yr ≈ AU$155/yr
Google Play Developer: US$25 one-time ≈ AU$40
ElevenLabs (voice clone + TTS): US$22/mo ≈ AU$35/mo
Privacy policy + ToS legal review: AU$500–1,500 (don't skip — mental-health adjacency)
Sentry, Capacitor, Stripe setup: free

Cost — ongoing variable

Apple IAP commission: 30% on in-app subscriptions (15% under Small Business Program if <$1M/yr — eligible). On $39.99/mo that's $12/mo skim, $59.70/yr on annual. Decision needed: web billing only, IAP-only (raise prices ~42%), or hybrid (recommended).
ElevenLabs Conversational minutes: ~6c/min. If users average 10 min/session × 20 sessions/mo = $12/user/mo cost on a $39.99/mo plan. Unit-economics watch-out.
Stripe: 1.75% + 30c (AU domestic) / 2.9% + 30c international.

Risk flags

High Voice coaching quality unproven. Needs 50+ private-beta conversations before launch. Single biggest risk to "no rework."
High Apple/Google review on health/therapy claims. Keep marketing language to skill-training ("read people, manage state"), avoid clinical claims. Category: Education, not Health & Fitness.
Med Liability on trauma-adjacent techniques (Fast Phobia Cure, Timeline Therapy, EMDR-style). Disclaimer alone isn't enough — needs in-app modals + role-boundary gating before each high-risk technique.

Project 03 — B2B SaaS

Tender Hunter → scales to 1,000

Stable production system. 10 → 100 → 1,000 customers without re-architecture.

Status

40%

Hours

~95h

Nights

11–15d

Cost

~$1.5k

What works today

7 AU portals scraping daily — AusTender (federal), Buy NSW, QTenders, Tenders WA, Buying Vic, SA Tenders, VendorPanel. 1,696 tenders ingested, 1,076 docs, 65% classified by Claude Sonnet 4.6. Daily cron at 06:00 AEST runs in 30 min. Match engine scores 100-pt across 7 dimensions (industry, work type, value, location, certs, clearances, remote). 1 user (you), Telegram alerts on ≥50% matches. 2 hand-written briefs prove the value-prop. SHIP_PLAN already drafted (8 phases, 77 tasks).

Gaps to close

No auth, no billing, no email. Three core production systems missing. Schema is multi-tenant ready, but no NextAuth, no Stripe webhooks, no transactional email.
No customer-facing dashboard. Match feed, profile editor, billing portal, feedback buttons all to build.
Brief generation isn't wired as API. Only manual briefs exist; need /api/brief/[match_id] Claude call + PDF render + tier gating.
SA scraper failing daily (Cloudflare/FlareSolverr fragility) and VIC scraper returns 0 (likely portal change).
No marketing site at tenderhunter.com.au — landing, pricing, FAQ, legal, the 70-chapter content guide.
No observability. No Sentry, no metrics, no alerting on cron miss, no GDPR export/delete endpoints.
No matcher feedback loop. Without "not relevant" / "saved" buttons, scoring stays static — churn at scale will hit 30–40%.

Effort — SHIP_PLAN phases

Phase	Focus	Hrs
P0	Hosting, DNS, Postgres role, Stripe test, Loops/SES, GitHub	3
P1	Marketing site (landing, pricing, legal, FAQ)	8
P2	Auth (NextAuth magic-link), signup, Stripe webhook, multi-tenant audit	10
P3	Email delivery + drip sequence	5
P4a	Guide content — 12 of 70 chapters (parallelisable)	25
P5	Dashboard MVP + feedback loop	12
P6	Brief generation API (Claude + PDF + tier gates)	8
P7	Hardening (security, load test, GDPR, monitoring)	8
+	Scraper fixes (SA, VIC), testing, Stripe live	16
Total		~95

Cost

Domains tenderhunter.com.au + tenderhunter.au: already held (sunk)
Hosting (Hetzner): zero marginal — already paid
Privacy/ToS legal review: AU$500–1,000
Loops/SES email: ~AU$30/mo at 1,000 customers
Claude classification at scale: AU$45–1,500/yr (cost passes through Predator tier)
Stripe Atlas, NextAuth, FlareSolverr: free

Pricing locked: Scout free / Hunter $79/mo / Predator $199/mo + $99 one-off briefs.

Already half-done

SHIP_PLAN is live at plan.illumen.dev/tenderhunter-ship awaiting your review. 5 minutes of YES/MODIFY/NO clicks unblocks ~95 hours of build. Worth doing this week.

Risk flags

High VendorPanel TOS explicitly prohibits commercial scraping. Largest portal (716 tenders). Mitigate by framing as "supplier alert tool" (not competitive intel) and pursue API partnership early.
Med SA Cloudflare breakage — fallback via AusTender federal feed when SA fails.
Med Matcher accuracy without feedback loop drives 30–40% churn in first 30 days. Phase 5 must ship before public signup.

Cold walk-in kit

Equipment for in-person sales

You self-resolved this correctly. Confirmed:

Laptop + S24 Ultra hotspot is sufficient. Light, fast, looks professional.
Add: USB-C → HDMI dongle (~AU$30) in case a prospect wants to throw the audit on their own monitor. Cheap insurance.
Tablet not needed. Save the AU$800–1,500.
Print 2–3 sample audit reports (~AU$50) to leave on the desk after you walk out. The lingering reminder closes deals later.

Total walk-in kit: <AU$100.

Legal & tax

Business structure recommendation

Short version

Keep all three under Illumen Group Pty Ltd for now. Do NOT spin up new entities yet. But set up a discretionary trust before any of these takes a paid customer — currently the highest-leverage thing missing.

Why no new entities now

Illumen audit isn't a product — it's an internal lead-gen tool. No separate structure needed, ever.
Tender Hunter is B2B SaaS with low liability exposure. Government tender alerts = informational service. Stays cleanly under Illumen Group.
NLP Mastery is the only one with real liability exposure (mental-health adjacent, consumer audience, App Store distribution). But at pre-revenue / pre-launch stage, separate entity adds cost ($500 setup + $332/yr ASIC) for marginal protection. Public liability + professional indemnity insurance ($500–1,500/yr) is much better value at this stage.

This month — priority order

Set up discretionary trust (~AU$1,500–2,500 with a tech-savvy lawyer). Trustee = Illumen Group Pty Ltd. Beneficiaries = you + future spouse + bucket company. Time-critical: must be in place before first paying client. Tender Hunter and NLP both about to hit revenue. Caveat: get legal advice on divorce timing — setting up trusts mid-divorce attracts scrutiny.
Public liability + professional indemnity insurance for Illumen Group covering all three product lines. ~AU$500–1,500/yr. Covers NLP user-harm risk, Tender Hunter data-accuracy claims, agency client disputes.
R&D Tax Incentive registration for FY26. The AI prospector and matcher likely qualify. 43.5% refundable offset = ATO writes a cheque even at zero revenue. Could be AU$5–15k against hosting + Anthropic + dev expenses. Find an R&D specialist on contingency.
Prepay 12 months hosting / Anthropic / coworking before June 30 — pulls FY27 expenses into FY26 deductions.

When to revisit

Spin NLP into its own Pty Ltd if/when: (a) MRR clears AU$10k/mo on the app specifically, OR (b) any user-harm claim surfaces, OR (c) outside investment specifically into the app.
Reconsider when leaving Australia. Non-resident structuring may obviate Australian Pty Ltd questions entirely. Don't over-build AU structure if exiting in <12 months.

Side by side

Comparison & priority

Project	Done	Hours	Cost	Priority
Illumen audit Cold-walk-in deals @ $2–3k	75%	24–32h 2 nights	<$100	#1 Fastest cash
Tender Hunter $79–$199/mo SaaS, recurring	40%	~95h 11–15 days	~$1.5k	#2 Plan ready
NLP Mastery $39.99/mo + $199/yr consumer	60%	350h full 150h web-only 15–20 / 6 days	~$2k yr 1	#3 Biggest upside, longest runway

Recommended sequencing

How to actually run this

This week — finish Illumen audit (24–32h). Cold-walk 20 Perth businesses → close 2–3 → fund everything else. Lowest risk, fastest cash.
Weeks 2–3 — ship Tender Hunter v1 per existing SHIP_PLAN. Recurring revenue starts.
Weeks 4–8 — NLP web-only launch first (150h). Validate voice coaching + spaced rep with paying users. Then invest the additional 200h to native iOS/Android wrap once unit economics are proven. Don't burn 350 hours and AU$2k on store fees before validating that voice coaching converts.

This sequence also matches the tax strategy — set up the trust before Tender Hunter takes its first dollar (week 2), insurance in place before NLP touches a paying user (week 4), R&D registration runs in parallel.

The single biggest unblock

Tender Hunter SHIP_PLAN is already drafted at plan.illumen.dev/tenderhunter-ship and awaiting your review. 5 minutes of YES/MODIFY/NO clicks unblocks ~95 hours of build. Do this before anything else.