Build Timeline
March 24–25
Days 1–2 — The First Scripts
tools
"I just want to make image ads faster."
tools
First Image Ad Generator
The one that started it all. A simple script that takes a product photo and renders ad text on top of it using Pillow. No AI imagery — just deterministic text composition with brand colors and fonts. The goal was embarrassingly simple: stop making ads by hand in Canva.
What it did
- Takes a lifestyle photo + product shot + copy
- Renders headline, subheadline, CTA with brand typography
- Outputs 3 formats: feed (4:5), story (9:16), square (1:1)
- Pure Pillow — zero AI hallucination, pixel-perfect control
Why this matters
This was the seed. It wasn't sophisticated — but it worked. And once you realize you can generate ads in seconds instead of hours, you immediately start thinking: "What else can I automate?"
"Build a Python script that takes a lifestyle product photo and renders ad text on top of it. Use Pillow. I need headline, subheadline, and CTA with Inter font. Output in 4:5, 9:16, and 1:1 formats. Use these brand colors."
tools
Photo Intelligence
Realized the generator needed to "understand" photos to place text well. Built a vision analysis tool using Gemini that categorizes lifestyle photos by scene type and maps them to the right messaging.
The turning point
This was the first sign the project was growing beyond "a script." I didn't plan to build photo intelligence — but the generator needed it, so I built it. That pattern would repeat every day for the next week.
"I have lifestyle photos of the product but I need to know which ones go with which part of the ad. Build a tool that uses Gemini vision to analyze each photo and categorize it — ritual shot, ingredients close-up, portability, etc. Cache the results."
tools
Ad Copy Generator + Gemini Renderer
The image generator was fast, but I was still writing copy by hand. Built an AI copy generator, then experimented with a second rendering path — sending photos + copy to Gemini's image model to bake text into the scene. Two approaches running in parallel.
Dual approach
- Path A (Pillow): Deterministic, zero hallucination, pixel-perfect but "stamped on"
- Path B (Gemini): AI renders text into the scene — more cohesive but riskier
- This dual-path thinking became a pattern: always test multiple approaches in parallel
"Build an ad copy generator that takes brand info and writes Meta ad copy. Validate against real ingredients, check banned words. Also build a second renderer that sends the photo + copy to Gemini to render the text into the scene."
March 26
Day 3 — Architecture Emerges
client
tools
"All this brand info is hardcoded everywhere. That's going to break."
client
Brand Configuration System
Hit a wall: product names, colors, compliance rules, and audience info were scattered across every script. Built a structured JSON config system so one source of truth feeds every generator. This was the first moment it stopped being "scripts" and started being "architecture."
product shot → brand config
sku catalog
Files created
- brand_config.json — Core identity, voice, positioning
- skus.json — Full product catalog with flavor stories
- avatars.json — Target audience personas with messaging hooks
- compliance.json — Banned words, restricted claims, preferred reframes
- channels.json — Channel-specific settings and rules
- social_proof.json — Real testimonials and data points
The shift
This was the accidental architecture moment. I didn't set out to build a "brand engine" — I just didn't want to update the same product name in 5 different files. But once the config existed, every new tool could plug into it. One change propagates everywhere.
"I keep hardcoding brand info in every script. Build a JSON config system — brand identity, products, target audiences, compliance rules, and channel settings. Then build a loader and context builder that assembles it all into a prompt-ready format."
tools
Template Extraction from Competitor Ads
Instead of designing ad layouts from scratch, fed competitor swipe-file ads to Gemini vision and extracted their layouts as reusable JSON templates. 16 proven ad layouts reverse-engineered into zone-based positioning.
How it works
- Feed a competitor ad image to Gemini
- Extract layout zones with positions (x, y, width, height as percentages)
- Output as reusable JSON template
- 16 templates extracted: product hero overlays, listicles, review cards, comparison ads
"I have a folder of competitor ads that perform well. Use Gemini vision to extract their layouts as JSON templates — headline position, subheadline position, CTA position, image zones. I want to reuse these layouts with our own brand content."
tools
Scene Compositor
The bridge between photo intelligence and ad rendering. Takes a scene photo + JSON template + copy data and composites the finished ad. Smart text placement with drop shadows, contrast optimization, auto-font sizing.
"Build a compositor that takes a scene photo, a JSON layout template, and generated copy, then composites the finished ad. It needs smart text rendering — drop shadows for contrast, auto-sizing fonts, and support for 3 ad formats."
March 27
Day 4 — The Knowledge Gap
knowledge
tools
"The ads look good... but the strategy behind them is shallow. I need real methodology."
tools
Transcriber Tool
The generators were producing ads, but the underlying strategy felt thin. Real practitioners share deep methodology in courses and YouTube videos. Built a modular CLI that downloads videos from YouTube and online course platforms, then transcribes them using Whisper.
What it does
- Downloads from YouTube (yt-dlp) and online course platforms (video extraction)
- Transcribes audio using OpenAI Whisper API ($0.006/min)
- Outputs clean markdown files organized by source
- Handles chunking for large files (25MB+ episodes)
Why this exists
I realized the generators were only as good as the thinking behind them. If I wanted the copy engine to write like a top media buyer, it needed to learn from top media buyers.
"Build a modular transcription CLI that can download videos from YouTube playlists and online course platforms, extract audio, and transcribe with Whisper. Support chunked transcription for large files. Output as organized markdown."
knowledge
First Knowledge Ingestion — Paid Ads Methodology
Transcribed 75 lessons from a top practitioner's paid ads course and YouTube channel. Account structure, creative testing frameworks, scaling playbooks. This became the backbone of every Meta Ads decision in the system.
The realization
This was the moment the project changed trajectory. Once I had real practitioner methodology in searchable markdown, I realized: this isn't just for my ads. This is the foundation for an entire agency's knowledge base. The scope started expanding.
"Transcribe all 50 lessons from this paid ads course. Use the platform downloader to extract the video sources, then transcribe with Whisper. Also grab the 25 most recent YouTube videos from the same creator."
Friday night dinner
That evening at Friday dinner, a family friend who's a developer asked what I'd been working on. I showed him Claude Code on the desktop app and he said: "Why aren't you using the terminal? You can run multiple sessions at once."
My mind broke a little. Up until that moment, I'd been seriously thinking about borrowing my mom's laptop just to get a second Claude Code window running. I didn't know you could open multiple terminal tabs and run Claude in each one.
Then he mentioned how he keeps hitting rate limits on the $100 Pro subscription. I was confused — we're on the same plan and I wasn't getting anywhere close to those numbers. That's when I realized how much faster he was working by running parallel sessions. He was getting 3x the output I was because he'd have Claude building one thing while he prompted another.
I went home that night, opened three terminal windows, and everything changed. My output roughly tripled overnight. The next four days — the marathon stretch where the real agency got built — wouldn't have been possible without that one conversation over dinner.
March 28
Day 5 — Knowledge Blitz
knowledge
"If I'm going to build methodology, I need to learn from the best across every channel."
knowledge
Multi-Channel Knowledge Expansion
Went on a transcription blitz across every marketing channel: email marketing flows and segmentation (39 lessons), Google Ads strategy and tactics (45 videos from two expert channels), offer building and customer acquisition frameworks (30 videos), and TikTok Shop playbooks from three top creators (59 videos). The knowledge base hit 248 transcripts.
Sources ingested
- Email marketing — 39 lessons: Klaviyo flows, segmentation, campaign cadence
- Google Ads strategy — 20 episodes: Performance Max, YouTube ads, attribution, CRO
- Google Ads tactical — 25 tutorials: account setup, PMax, Shopping, Search, Demand Gen
- Offer building — 30 videos: pricing psychology, acquisition models, business design
- TikTok operations — 59 videos across 3 creators: TikTok Shop, GMV ads, affiliates, LIVE selling
Cost
Total transcription cost for all 248 videos: ~$7.20 via Whisper API. Less than a coffee for an entire agency's knowledge base.
"Transcribe these YouTube playlists — filter for videos about Google Ads, PMax, and attribution from 2023 onward. Skip anything older or off-topic. Use chunked transcription for episodes over 25MB."
March 29
Day 6 — The Big Build
agency
"I have the knowledge, I have the tools. Time to build the whole operating system."
agency
Full 56-File Agency Infrastructure
The single biggest build day. Created the entire Wild Ventures operating system from a 200KB master architecture document: 56 files across 14 sections covering every aspect of running a performance marketing agency. Methodology, playbooks, SOPs, templates, legal docs — everything. This was the day the side project became a company.
All 14 sections built
- Methodology (6) — Creative framework, campaign architecture, performance loop, testing protocol, offer framework, channel strategy
- Playbooks (7) — Meta, Google, Native, TikTok, Snap, Email, SMS
- Copy Engine (5) — Ad copy, video scripts, email/SMS, landing pages, style guide
- Intelligence (5) — Creative taxonomy, performance tracker, report prompt, attribution
- Onboarding (5) — Intake form, brand bible, project setup, checklist
- Business (10) — Pricing, client selection, revenue model, 5 legal templates
- + Strategy Engine, Production, Funnels, Reporting, Lifecycle, Deployment, Knowledge Sources
"Here's our master architecture document. Build every file in it — all 14 sections. Reference the transcribed course knowledge for real methodology. Each file should be production-ready. Build the entire agency operating system."
tools
Generator Suite Expansion
Expanded the generator suite to cover every ad format: multiple rendering versions testing different approaches, endcard generators for mobile app ads, native formats, carousels, captions, and content calendars.
endcard · benefit stack
endcard · offer forward
endcard · studio clean
Generators built
- Ad generators v9/v10/v11 — Template+Pillow vs AI rendering, testing in parallel
- Endcard generators — AppLovin 9:16 format for mobile
- Native ad generator — Taboola/Outbrain formats
- Carousel, caption, and content calendar generators
"Build endcard generators for AppLovin 9:16 format, a native ad generator for Taboola/Outbrain, and carousel generators. Use the same brand engine pattern — load config, build context, generate."
tools
Deployment CLI + Meta API
Built the campaign deployment pipeline: a CLI that takes a campaign config JSON and pushes it live to Meta Ads via the Marketing API. Plus deployers for Taboola, Outbrain, and organic social. Dry run tested.
"Build a deployment CLI that takes a campaign config JSON and deploys it to Meta Ads. Use the Marketing API. Support dry-run mode. Also build deployers for Taboola, Outbrain, and organic social."
March 30
Day 7 — Platform + AI Video
platform
ai
client
"These tools are too good to live in a terminal. They need a platform."
platform
Full Platform Deploy — Railway + Netlify + Supabase
Deployed the complete SaaS platform: FastAPI backend on Railway (30 API endpoints), Next.js frontend on Netlify (7 pages), and Supabase PostgreSQL database (11 tables). Auth, onboarding wizard, creative dashboard — the full stack, live in production.
Backend (FastAPI on Railway)
- 30 API endpoints across 18 route files
- Services: ad generation, brand management, campaigns, creative, deployment, intelligence, media buying, reporting
- Supabase Auth + webhook endpoints for Meta events
Database (Supabase PostgreSQL)
- 11 tables: users, brands, assets, creatives, generations, campaigns, connections, quizzes, questions, result_buckets, leads
"Build the SaaS platform. FastAPI backend with Supabase for auth and database. Next.js frontend with Tailwind. I need: onboarding wizard, brand management, creative generation, campaign management, and a dashboard. Deploy backend to Railway, frontend to Netlify."
tools
Flagship Ad Generator v4 + Advertorial Engine
Unified all generator learnings into the flagship: 8 visual styles, 16 scene contexts, 6 color directions, 5 lighting moods — 3,840 unique combinations per product per audience. Then built the advertorial engine: full HTML presell pages with personal journey stories, listicles, and product reviews.
v4 · ugc raw · 9:16
v4 · editorial · 4:5
v4 · close-up macro
v10 · composited
v4 scale
8 visual styles × 16 scenes × 6 colors × 5 lighting moods = 3,840 unique combinations per SKU per avatar. Creative testing at a scale no human team can match.
Advertorial templates
- Personal journey — "How I discovered..." narrative arc
- Listicle — "7 reasons..." format for native platforms
- Product review — Expert-style analysis and recommendation
- Before/after — Transformation story with visual proof
- Newsroom — Editorial-style product coverage
"Build the unified ad generator. Combine the best of v9 (Pillow precision) and v4 (Gemini cohesion). Support 8 visual styles, 16 scene contexts, 6 color directions, 5 lighting moods. Then build an advertorial generator for full HTML presell pages."
ai
AI Video Engine
Integrated AI video generation for UGC-style ad videos. Hybrid pipeline: AI generates video clips for hook/body/reveal scenes, CTAs stay as static renders. Cost: $5-7 per 30-second video vs. $100-150 at agencies.
Architecture
- Hybrid pipeline: AI clips for dynamic scenes, static renders for CTAs
- Text overlay compositing on top of AI-generated footage
- Standard mode (~$0.07/s) and Pro mode (~$0.14/s)
"Add an AI video provider to the video engine. Hybrid approach — AI generates hook, body, and reveal scenes, CTAs stay as static renders. Composite text overlays on top."
client
Second Client Brand — Different Niche
Built a complete brand package for a digital course business in a completely different vertical. Full brand context, generators, niche intelligence from expert transcripts, landing page, email sequence, campaign config. Same tools, different niche, zero code changes. Proof the system scales.
Why this matters
Went from zero to launch-ready for a completely different niche in a single session. That's the power of the brand engine — plug in new configs, get a fully operational brand.
"Build a complete brand package for a new client in a totally different niche. Research the niche by transcribing expert YouTube channels, build niche intelligence, then create the full brand config, generators, landing page, email sequence, and campaign config."
knowledge
Video Production Knowledge Base
Transcribed 114 videos across 10 creators covering video editing, hooks, motion graphics, color grading, and sound design. Built the knowledge layer that informs the video engine's creative decisions.
"Transcribe videos from these 10 YouTube channels focused on video production. Filter for editing techniques, hooks, motion graphics, color grading, and sound design."
March 31
Day 8 — Creative Studio + Lead Gen
ai
deploy
"Clients shouldn't need a terminal. They need a studio."
ai
Unified AI Creative Studio
Built a unified creative generation system — one API aggregating access to multiple image models (Midjourney, Flux, 4o) and video models (Veo, Runway). Provider abstraction with fallback. Full Creative Studio UI: describe what you want, get an estimate, generate, review.
Architecture
- Provider abstraction layer — swap AI providers without changing application code
- Unified API for both image and video generation
- CreativeBrief model — captures client intent before generation
- Studio API — 7 endpoints for the full brief lifecycle
- Studio UI — brief builder, progress tracking, review gallery
"Build a Creative Studio into the platform portal. Use a unified AI API as the provider — it aggregates models like Midjourney, Flux, and video models. Build a provider abstraction layer. Add a brief-based workflow: describe what you want, estimate cost, generate, review."
ai
AI Creative Orchestrator
Added Claude-powered pipeline planning. Given a creative brief, Claude designs a multi-step pipeline: generate images → animate to video → generate voiceover → write copy → assemble final. Each step feeds outputs forward. The AI plans the AI.
Why this matters
Image-to-video always beats text-to-video for ads. The orchestrator knows this and automatically chains the optimal sequence. Clients just describe what they want — the system figures out how.
"Add an AI orchestrator that uses Claude to plan multi-step creative pipelines from briefs. A brief for a product video should plan: generate product images → animate to video → add voiceover → generate copy → assemble. Show real-time progress."
deploy
Marketing Eval Lead Magnet
Built a free marketing health scan tool — paste any URL, get a scored report card covering performance, SEO, backlinks, ad presence. Feeds into the discovery funnel to generate leads automatically.
Funnel architecture
/eval/ (free scan, shareable) → /discover/ (7-step brand questionnaire) → onboarding → book a call. Top of funnel is completely self-serve and viral.
"Build a marketing evaluation tool. Users paste a URL and get a scored marketing health report — performance, SEO, backlinks, ad presence. Score each section, give an overall grade, and funnel them to the discovery questionnaire."
platform
Cloud Storage Migration
Wired all generators to upload outputs directly to Supabase Storage instead of local filesystem. Migrated existing assets. Everything now served from CDN.
"Wire all the generators to upload outputs to Supabase Storage instead of saving locally. Migrate existing assets. Update the frontend to load from CDN URLs."
April 1
Day 9 — Ship It
deploy
"Everything works. Time to put it in front of people."
deploy
Creative Studio Production Deploy
Deployed the full Creative Studio to production — Railway backend with studio routes, Netlify frontend with the brief builder and review gallery. Live and client-accessible.
"Deploy the Creative Studio to production. Push backend to Railway, build and deploy frontend to Netlify. Make sure the model names show friendly labels."
deploy
Full Marketing Site + Internal Hub
Deployed the complete web presence: agency landing page, investor pitch deck with interactive valuation model, tech architecture overview, discovery questionnaire, demo galleries with all generated assets, and the internal team hub. Wild Ventures went from "a guy with scripts" to a company with a website.
demo gallery · v9 no-text
demo gallery · scene photo
advertorial · hero image
Pages deployed
- Landing page — agency positioning and capabilities
- Pitch deck — 14-slide investor presentation
- Valuation model — interactive ARR projections and cap table
- Tech architecture — codebase overview, integrations, roadmap
- Discovery questionnaire — 7-step brand intake
- Demo galleries — generated ads, endcards, videos, landing pages
- Internal hub — team operations, client materials, investor links
April 2
Day 10 — The Meta Layer
meta
"Let's document how we got here — so others can do it too."
deploy
This Build Journal
You're looking at it. A complete record of every build day, what was created, why it matters, and how we prompted Claude Code to build it. Using the tool to document the tool. Ten days ago this was a guy in a terminal making image ads. Today it's a full-stack AI-powered marketing agency with a platform, knowledge base, and client portal.
Why we built this
- Internal reference — look back and see how everything connects
- Marketing proof — show prospects the speed and depth of what AI + intention can build
- Educational content — teach others how to use Claude Code for real projects
- Inspiration — because the best things start as something small you didn't plan
"Build a visual timeline of our entire build journey. Start from Day 1 when I was just trying to make ads faster. Show how each day's builds led to the next. Include time worked, real output examples, and sample prompts so people can learn to use Claude Code themselves."
April 4
Day 12 — Real Deployment + The To-Do App
deploy
platform
"If we can generate ads, we need to be able to actually deploy them."
deploy
Meta Ads Real Deployment
Wired the Facebook Business SDK into the platform so generated creatives can be deployed as real Meta ad campaigns — not just exported files. Campaign creation, ad set configuration, and creative upload all happen through the API. Also built out Google and TikTok deployers with full campaign creation support.
What was wired
- Facebook Business SDK integration — real API calls for campaign + ad set + creative
- Google Ads deployer — campaign creation with keyword targeting
- TikTok deployer — full campaign lifecycle
- Row-level security on all deployment tables
platform
Do — To-Do Command Center
Built a full interactive to-do application at wildventures.ai/do. AI-powered chat assistant, Supabase auth, PWA with install support, journal mode, voice input, theme customization, haptic feedback, and sound effects. A complete productivity app that shipped in a single session.
Features shipped
- AI chat assistant built into the task manager
- Full PWA — installable on mobile with custom icons
- Haptic feedback and sound effects for interactions
- Journal mode for daily reflections
- Voice input for hands-free task creation
- Multiple themes with smooth transitions
ai
Creative DNA + Intelligence Loop
Built a unified metadata tagging system for every creative asset — hooks, visual style, emotional tone, audience segment. This "Creative DNA" feeds the intelligence analytics layer that tracks what's working and why. Wired the intelligence loop into the client portal chat so insights surface automatically.
The intelligence layer
- Creative DNA — structured tags for every asset (hook type, tone, visual style)
- Intelligence analytics dashboard
- Attribution, credits, and tracking services
- Portal chat integration — AI references performance data in conversations
deploy
Vision Pages + Pitch Deck Rewrite
Added three interactive vision pages showing the 5-phase evolution from agency to platform to marketplace. Rewrote the entire pitch deck to align with the business creation engine vision. Separated internal and external hubs so team tools and public marketing have clean boundaries.
Pages shipped
- Strategic overview document
- 3 interactive vision pages
- Rewritten pitch deck — marketplace/creation engine angle
- Separate internal hub (team) and external hub (public)
April 5
Day 13 — Full-Stack Polish
platform
deploy
"Time to make everything actually work end-to-end."
platform
Campaign Orchestration Layer
Built the orchestration layer that coordinates multi-channel campaign launches. Added Snapchat and Klaviyo deployers to the existing Meta, Google, and TikTok channels. All five channels wired into a single deployment service. Shipped a Launch page UI for one-click multi-channel deploys.
Channels wired
- Meta (Facebook/Instagram) — via Business SDK
- Google Ads — campaign + keyword targeting
- TikTok — full campaign lifecycle
- Snapchat — new deployer
- Klaviyo — email campaign deployer
- Launch page — multi-channel deployment UI
deploy
Public AI Chat Widget
Added an AI chat widget to every external page on wildventures.ai. Visitors can ask about the platform, services, and methodology. Went through five iterations of UI polish — font, padding, bubble sizing — then wired real features: loading indicators, conversation deletion, auto-polling for generation status, and conversation logging to the database.
Chat iterations
- v1 — Basic widget on all pages, Claude Sonnet backend
- v2-v4 — Inter font, refined padding/radius/sizing
- Loading indicators, delete chats, auto-poll generations
- All conversations logged to database for review
- Fixed JS string literal bug from single quotes in font-family
platform
Generation Pipeline Fixes
Deep debugging session on the creative generation pipeline. Wired Gemini as the v4 image engine for all AI image generation. Fixed native ads and carousels to use the real generation pipeline. Fixed creative downloads, preview rendering, and SKU-aware product shot lookup. The platform now generates, previews, and downloads every asset type correctly.
Fixes shipped
- Gemini v4 image engine wired for all generation
- Native ads + carousels → real generation pipeline
- Creative download triggers file save (not new tab)
- Smart previews — correct rendering for every asset type
- SKU-aware product shot lookup for image generation
- Dockerfile moved to repo root for Railway deploy
deploy
Immersive Landing Homepage
Replaced the hub page with a fully immersive landing homepage. Added legal pages (privacy policy, terms of service, data deletion) and a 1024px app icon. Fixed home links across all sub-pages so the Wild Ventures logo navigates consistently.
Pages shipped
- New immersive landing homepage
- Privacy policy + terms of service + data deletion page
- Consistent logo → home navigation across all pages
April 6
Day 14 — Client Onboarding Pipeline
client
platform
"We need anyone to be able to walk in and become a client — no manual setup."
platform
Client Onboarding Pipeline
Built the full automated onboarding flow: a public intake form captures brand details, which triggers account creation, RAG knowledge ingest, and portal provisioning. A new client goes from "never heard of us" to "logged into their portal with AI that knows their brand" — without anyone on our side lifting a finger.
Pipeline stages
- Public intake form — captures brand info, goals, assets
- Auto account creation — Supabase auth + brand record
- RAG ingest — brand knowledge indexed for AI context
- Portal provisioning — client gets instant access
deploy
Discover + Plan Pages + Team Matching
Merged the discover and plan pages into a single cohesive flow. Added team-to-brand matching so the right specialists get routed to the right clients based on skill alignment and availability.
What shipped
- Combined discover + plan into one seamless page
- Team brand matching algorithm
client
Peptide Potion — New Client Brand
Onboarded a new DTC skincare brand: Peptide Potion (GHK-Cu serum, $109). Built their product detail page and drafted an onboarding brief for founder review. Second real client brand in the system after VitaBar.
Deliverables
- Product detail page (PDP) — full copy, layout, brand styling
- Onboarding brief drafted for founder review
- Brand added to clients/ directory