Master Strategy & Technical Map

Affirmology AI
How to Build It

A full architecture for personalized affirmational audio — from data accuracy to agent pipelines to ads — with a realistic sprint plan for a solo founder with limited time.

Gene Keys First Agent-Driven No Full-Time Required Yet Bootstrap Path
01 — The Data Accuracy Crisis

Why AI Gets Charts Wrong — And How to Fix It

The critical flaw: ChatGPT and Claude hallucinate astrological and Gene Keys data. They confidently produce wrong chart placements, incorrect sphere assignments, and fabricated line meanings. This is a trust-destroying problem for a product that must feel intimate and accurate. The fix is not prompting harder — it's architecture.
// WHY AI GETS IT WRONG → FIX ARCHITECTURE
❌ What's Happening Now
LLM stores no real ephemeris data
Invents planet positions from pattern-matching
Confidently wrong Gene Key assignments, House cusps, Nodal axis
✓ The Fix: Separation of Concerns
Calculation layer: Swiss Ephemeris / Astro-Seek API → NEVER ask LLM to calculate
Structured JSON chart data → fed INTO the LLM as ground truth
LLM only does: interpretation, affirmation writing, tone — never calculation
// GENE KEYS SPECIFIC FIX
Birth Date + Time + Place
Swiss Ephemeris
calculates exact
planetary degrees
Degree → I Ching
Hexagram lookup
(static mapping table)
Hexagram = Gene Key
(1–64, deterministic)
Line # from
degree sub-division
★ The Gene Keys sequence (Life's Work, Evolution, Radiance, Purpose, etc.) maps directly to specific planetary positions. This is mathematical, not interpretive — it must be computed, not guessed.
Key principle: Build a Chart Data API layer that returns verified JSON for any birth data. Every downstream agent receives this verified JSON — no agent ever re-derives chart data from scratch. The LLM's job is purely creative: interpretation → affirmation generation → script writing.
02 — Core Production Pipeline

From Birth Data to Audio Track

📋
User Input
Birth date
Birth time
Birth city
Preferences
Life themes
🔭
Chart Engine
Swiss Ephemeris
AstroSeek API
Gene Keys mapper
HD Bodygraph
→ verified JSON
📚
Knowledge RAG
Gene Keys DB
Astro interpretations
HD type meanings
Affirmation bank
Vector search
✍️
Script Agent
Claude Sonnet
Affirmation builder
I AM statements
Hook + flow
QA check
🎙️
Voice Render
ElevenLabs API
Voice selection
Pacing / pauses
SSML control
MP3 output
🎵
Music Layer
Suno / Udio
Epidemic Sound
Custom beds
FFmpeg mixing
Mastering
📱
Delivery
Web app / PWA
Secure stream
User library
No iTunes leak
Feedback loop
// KNOWLEDGE DATABASE ARCHITECTURE — The Bank That Grows
Gene Keys Knowledge Bank
  • 64 Keys shadow/gift/siddhi
  • 6 Lines per key variant
  • 12 Spheres role meanings
  • Codon rings groupings
  • Public YouTube scraped + summarized
  • NotebookLM research base
Affirmation Bank
  • Per-Key templates gift-level focus
  • Per-Sphere templates role-specific
  • Transit templates new/full moon
  • Scored variants A/B tested
  • User feedback auto-improves
Vector Store (RAG)
  • Pinecone / Chroma embeddings
  • Semantic search on user config
  • Context injection into LLM prompt
  • Grows with use self-improving
QA & Self-Improvement
  • Output scoring agent pre-delivery
  • User ratings in-app
  • Weak affirmations flagged + rewritten
  • Bank improves every cycle
03 — Agent Architecture

The Multi-Agent System

Rather than one monolithic AI call, Affirmology uses specialized agents — each with a narrow job, high-quality prompt, and ability to verify its own output. The orchestrator coordinates them. This dramatically improves reliability and makes the system upgradeable piece by piece.
// AGENT MAP — GENE KEYS AUDIO PRODUCTION
🎯 Orchestrator Agent
Receives user input. Sequences all other agents. Manages state. Returns final audio job.
🔭 Chart Agent
Calls Swiss Ephemeris. Returns verified JSON. Gene Keys sequence. HD type. NEVER hallucinates — pure computation.
📚 Research Agent
Takes chart JSON. Queries vector DB for relevant Gene Key meanings, sphere roles, line nuances. Returns context bundle.
✍️ Script Agent
Takes context bundle. Writes affirmational script. I AM statements. Hooks. Repetition patterns. Checks tone + flow.
🔍 QA Agent
Reviews script for generic language, inaccuracies, weak hooks. Scores it. Sends back for revision if below threshold.
🎵 Audio Agent
Sends approved script to ElevenLabs. Selects music bed. Mixes via FFmpeg. Returns final MP3 to orchestrator.
// MONTHLY TRANSIT VARIANT — Additional Agents
🌙 Transit Agent
Pulls current ephemeris data. Identifies active transits. Calculates how they aspect user's natal chart. Returns transit JSON.
🗺️ Relevance Agent
Cross-references transit with user's Gene Keys sequence, HD channels, natal placements. Ranks significance. Filters noise.
📝 Update Script Agent
Writes transit-specific affirmations layered on top of user's baseline profile. Seasonal + personal blend.
04 — Infrastructure: Mac Mini vs Cloud

Claude Ecosystem vs. Local Agents — A Real Comparison

You're asking the right question early. Running agents on a Mac Mini (especially M-series) is genuinely viable for this use case — but there are real tradeoffs. Here's the honest comparison.
Dimension
☁️ Claude API + Cloud Stack
Anthropic API + managed services
🖥️ Mac Mini Local Agents
Ollama / local LLMs on M-series
Setup Complexity
Low — API keys, no infra management
Medium — model downloads, local server config, port management
Intelligence Quality
Highest — Claude Sonnet 4 / GPT-4o frontier models. Best affirmation writing quality.
Good for basic tasks. Llama 3 / Mistral solid but not frontier. Affirmation creativity is noticeably weaker.
Cost at Scale
Per-token costs add up. ~$0.05–0.15 per full audio script generation. Manageable at low volume, watch at scale.
Near-zero marginal cost once hardware is paid. Mac Mini M4 Pro ≈ $1,400 one-time.
Latency
Fast API response. Production-ready.
Slower on 7B–13B models. M4 Pro handles 70B quantized reasonably but not instantaneously.
Privacy / Data
Birth data sent to Anthropic/OpenAI. Must address in privacy policy. Generally acceptable.
All data stays local. Strong privacy story — meaningful for spiritual audience.
Reliability
99.9% uptime SLA. Auto-scales.
Single point of failure. Your internet goes out, service is down. No redundancy without extra engineering.
Audio Tools (ElevenLabs)
ElevenLabs is cloud-only — always API regardless of local vs cloud LLM
Same — ElevenLabs always cloud API
Best For
Demo → launch → scaling. Fastest path to quality product.
Research phase, batch processing affirmation banks, cost optimization once patterns are proven.
Claude Code / Agents
Claude Code runs locally on your Mac but calls Anthropic API. Best of both — local control, frontier intelligence.
Ollama + LangGraph local loop. Works well for experimentation. Lower quality ceiling.
Recommendation: Use the Claude API + cloud stack for production (quality matters for your audience). Use a Mac Mini with Ollama/Claude Code for batch research, affirmation bank building, and cost-free experimentation. Hybrid is the answer — not either/or. A Mac Mini M4 Pro as a local dev/batch server + Claude API for live user requests is the smart setup.
// RECOMMENDED HYBRID ARCHITECTURE
Mac Mini
(local research + batch)
Claude Code CLI
(orchestration scripts)
Anthropic API
(live script generation)
ElevenLabs API
(voice render)
Supabase
(user data + audio store)
Web App / PWA
(delivery)
05 — Ads & Marketing AI System

Automated Creative → Test → Scale

Step 1
🎯
Audience Research
Claude / Perplexity
Gene Keys Reddit
FB group analysis
Competitor audits
Step 2
✍️
Copy Generation
Claude writes 10+
hook variants per
audience segment.
Pain → transformation
Step 3
🖼️
Visual Creation
Midjourney / Flux
Cosmic, celestial
Canva for text overlay
CapCut for video
Step 4
🎬
Demo Clips
30-sec audio demo
Show the FEELING
Waveform visuals
Testimonial overlay
Step 5
🧪
Nano Testing
$5–10/day per ad
Meta + TikTok
5–10 variants
Hook A/B test
Step 6
📊
AI Analysis
Claude reads results
Identifies winners
Flags weak angles
Suggests next test
Step 7
🚀
Scale Winners
Winning creative → $50+/day
Lookalike audiences
Influencer seeding
Retargeting layer
Influencer Strategy
  • Gene Keys community Facebook groups
  • HD YouTube creators free sample offer
  • Astrology TikTok micro-influencers
  • Give personalized audio → authentic review
Funnel Structure
  • Free sample audio hook + email capture
  • Email nurture 3–5 day sequence
  • Starter Kit offer $47–97 one-time
  • Monthly upsell transit membership
Email & CRM Tools
  • ConvertKit / Beehiiv email sequences
  • Claude writes emails per segment
  • Monthly transit drops auto-sent
  • Lemon Squeezy payments
AI Ad Tools Stack
  • AdCreative.ai creative testing
  • Foreplay.co competitor ad spy
  • Make.com ad automation flows
  • Claude copy + analysis
06 — Distribution

With App vs. Without App

📱 With App (PWA or Native)
Pros
  • Users stay in your ecosystem — not iTunes
  • Push notifications for new transit drops
  • In-app feedback and ratings loop
  • Personal library feels premium
  • Subscription billing native experience
  • Analytics on listening behavior
Cons / Costs
  • More build time (6–12 weeks minimum)
  • Apple App Store 30% cut on in-app purchases
  • Ongoing maintenance overhead
  • Higher dev cost if outsourced
Best for: Post-demo, post-100-customer phase. PWA first (skip App Store), native later.
🌐 Without App (Web + Email)
Pros
  • Ship in days not months
  • Simple: web form → Stripe → email delivery
  • No App Store fees or review delays
  • Frictionless purchase flow
  • Test product-market fit fast
  • Works perfectly for demo + Starter Kit phase
Cons
  • Users download files → can share freely
  • No engagement/retention loop
  • No push notification for new drops
  • Harder to build listening habit
Best for: Sprint 1 demo + first 50 customers. Validate everything before building the app.
Staged approach: Launch with a simple web form + Gumroad/Lemon Squeezy for payments + email delivery of MP3s. Once you hit 50 paying customers and know what they love, build the PWA. This avoids wasting months building an app for a product still being validated.
07 — Sprint Plan (Part-Time Founder Reality)

What to Actually Do and When

You have a relationship coaching business to launch first. These sprints assume 5–10 hours/week of spare time now, with an option to go hard in a few months. Gene Keys is the proving ground.
Sprint 1
The Undeniable Demo
4–6 weeks · 5 hrs/week
  • Build Gene Keys chart accuracy layer: Swiss Ephemeris + static hexagram→Gene Key map in Python
  • Create NotebookLM knowledge base: 64 keys, 6 lines, 12 spheres using public YouTube sources, not Rudd's books
  • Write 10 high-quality affirmation templates for Gene Keys (shadow release, gift activation, siddhi contemplation)
  • Build one end-to-end flow manually: take a real chart → pull context → write script → ElevenLabs → music mix
  • Produce 3 demo tracks for 3 real people. Get feedback. Does it feel personal? Magical? Do they replay it?
Sprint 2
Simple Launch Infrastructure
3–4 weeks · 5 hrs/week
  • Build intake form: Typeform or simple web form collecting birth data + preferences
  • Semi-automated pipeline: form submission → triggers Claude agent → ElevenLabs → MP3 → email delivery
  • Set up Lemon Squeezy for Starter Kit payment ($67 test price)
  • Landing page: ConvertKit or Carrd — demo clip front and center
  • Soft launch to warm network: Gene Keys Facebook groups, HD communities. 10 paid customers goal.
Sprint 3
Subscription & Transit Engine
4–6 weeks · go-hard mode
  • Build transit calculation layer: current ephemeris → aspect analysis vs natal chart
  • Monthly audio generation pipeline: triggered by new/full moon dates
  • Add Western astrology layer alongside Gene Keys
  • Stripe recurring billing + simple member portal (Memberstack or Outseta)
  • Affirmation bank QA loop: score outputs, flag weak statements, auto-improve
Sprint 4
PWA App + Ads Scale
6–8 weeks · with capital or contractor
  • Build PWA: user auth, personal audio library, in-app streaming, push notifications
  • Launch paid ad campaigns: Meta + TikTok. $20/day nano-test budget to start.
  • Influencer seeding program: 20 free audios to Gene Keys / HD creators
  • Music vibe selector: 4 aesthetic options per audio
  • Prepare investor deck with real traction data. This is when the pitch is credible.
08 — Future Horizon

How This Compound Gets Bigger

NOW
Sprint 1: Gene Keys Demo
Manual pipeline, 3 demo tracks, validate the magic. This is the proof of concept. If 3 people replay their audios, you have a product.
Sprint 2–3: First Revenue
50 paying Starter Kit customers. Monthly transit subscription tested. Real feedback loop. Affirmation bank starts self-improving.
6 MO
AI Music Will Be Radically Better
Suno, Udio, and new entrants will offer custom stem control, key/tempo locking, and mood-to-music prompting far beyond today. Music beds will be fully generative and personalized to the user's vibe preference — not template-based. Build the pipeline to be music-source-agnostic now so you can swap in new tools effortlessly.
User Voice Integration
ElevenLabs voice cloning will be accessible enough to let users record their own voice as the narrator. Hearing your own voice speak your affirmations is a psychologically powerful upgrade. This becomes a premium tier feature.
12 MO
Full System Expansion
Add Vedic astrology, Human Design deep channels, numerology. The knowledge bank is now a full multi-system oracle. Each new system = new content line = new revenue. The infrastructure built in Sprints 1–3 handles all of it with new data plugged in.
Community & Network Effects
Users sharing their audios, comparing Gene Keys sequences, discussing transit impacts. The product becomes a community touchpoint — not just a file delivery service. This is where the moat deepens. Peer discovery inside the app: "Others with Gene Key 36 in their Life's Work sphere also loved..."
B2B / Practitioner Channel
Gene Keys facilitators, HD analysts, and astrologers can white-label Affirmology audios for their clients. A practitioner pays a flat fee and resells personalized audios as part of their service. Entirely new revenue stream requiring almost no new product work.
The Path Forward Next sessions: Market research → Financial model → Prototype build → Investor deck.
The demo is the pitch. Build the demo first. Everything else follows.