01 The Challenge
Brand Inputs Live Everywhere: Websites, Decks, Drives. Producing On-Brand Creative For Every Social Format Took Marketing Teams Days Of Back-And-Forth.
Case Study · 01 / AI Marketing Platform
Tessa
An AI Marketing Platform That Learns A Brand From Its Website, Builds A Reusable Brand DNA, And Turns One Brief Into On-Brand Ad Creatives For Every Channel, In Minutes.
Product Design & Full-Stack Engineering ·
- 01 · Product
- Tessa
- 02 · Category
- AI MarTech
- 03 · Timeline
- 8 Weeks · 2026
- 04 · Scope
- Design + Full-Stack
- 05 · Platforms
- Web App · API
01 · By The Numbers
The Build,
By The Numbers
Counted Straight From The Codebase And Git History. No Vanity Metrics, No Projections.
29
Ad Themes
Product Hero To UGC Style
6
Creative Formats
IG · FB · X · LinkedIn · Hero
4
Worker Queues
Brand · Campaign · GPU · Default
132
Commits
Feb → Apr 2026
02 · Overview
From One Brief
To Every Format
02 Our Approach
We Scrape The Site Into A Structured Brand DNA, Then Fan One Brief Out To Queued Gemini Workers That Write Copy, Pick Ad Themes And Render Each Format.
03 The Outcome
Marketers Go From A URL To A Library Of Brand-Consistent Ads, Then Refine Any Creative By Chat Or AI Image Edits, Metered By Simple Credits.
03 · What We Built
Five Engines,
One Brand DNA
Every Generation Reads From The Same Brand Profile, So Copy, Layout And Imagery Stay On-Brand Across Channels.
Brand DNA Extraction
BrowserBase Screenshots And Gemini Vision Distil Voice, Visuals, Logo And Products Into A Reusable Profile, Exportable As A Brand PDF.
Campaign Engine
One Brief Fans Out To Every Selected Format, With Copy, Theme And Layout Chosen Per Channel.
29 Ad Themes
From Product Hero And Social Proof To UGC And Meme-Style Layouts, Parsed Into Renderable Templates.
Chat & Image Editing
Refine Any Creative In Conversation Or Edit Imagery With AI, Every Change Kept In The Campaign Thread.
Credits & Billing
Usage-Based Credits For Scrapes, Formats And Edits, With Checkout, Webhooks And Product Analytics.
04 · Architecture
Queue Everything,
Poll The Result
A FastAPI Core Hands Every Heavy Job To Celery Workers. Clients Poll Task Status While Gemini, BrowserBase And The Renderer Do The Work.
- Sync · REST / GraphQL
- Async · Kafka Events
- Service Node
~50 Endpoints · 4 Queues · AWS ECS Fargate
Read The Architecture As Text
- Clients
- Dashboard (Next.js 16 · React 19), Onboarding (Brand Setup Flow), Public API (Per-User API Keys)
- Edge
- FastAPI (~50 Routes · Rate Limits), Auth (Clerk · RLS Context)
- Services
- Brand Workflow (Scrape · Vision · PDF), Campaign Workflow (Prompt · Copy · Themes), Render Service (Layouts · CairoSVG), Product Catalog (Pages · Reviews), Celery Workers (Brand · Campaign · GPU)
- Data
- PostgreSQL 16 (14 Tables · RLS), Redis 7 (Broker · Cache), Amazon S3 (Assets · Logos), Gemini (Text · Vision · Image)
05 · Stack
The Stack,
And The Code
A Typed Next.js Front End Over A Python Pipeline, Picked So The GPU-Heavy Work Can Scale Independently Of The API.
- 01 Frontend
- Next.js 16
- React 19
- TypeScript
- Tailwind v4
- 02 Growth
- Clerk
- PostHog
- Dodo Payments
- 03 Backend
- FastAPI
- Python 3.12
- Celery
- Pydantic v2
- 04 Data
- PostgreSQL 16
- SQLAlchemy 2
- Redis 7
- Amazon S3
- 05 AI
- Gemini 3 Flash
- Gemini 3 Pro Image
- BrowserBase
- Playwright
- 06 Infra
- AWS ECS Fargate
- RDS
- Docker
- GitHub Actions
async def generate_campaign( user_id: str, creative_types: List[str], brief: str, brand_dna: Optional[Dict] = None, user_prompt: Optional[str] = None, dna_id: Optional[str] = None, selected_image_urls: Optional[List[str]] = None, selected_product_context: Optional[Dict[str, Any]] = None, layout_options: Optional[Dict[str, Any]] = None, brand_owner_id: Optional[str] = None, ) -> Dict[str, Any]: """Runs the multi-asset marketing campaign generation pipeline.""" # Whoever owns the brand is the lookup owner; default to caller effective_brand_owner = brand_owner_id or user_id agent = MarketingAgent(user_id=user_id) async with AsyncSessionLocal() as db: await set_rls_context(db, user_id)$ celery -A worker worker -Q campaign,gpu→ campaign + gpu queues consuming06 · PageSpeed
Fast, Accessible,
Found In Search
Google PageSpeed Insights Scores For tessa.so: Speed, Accessibility And SEO Engineered In From Day One.
- 96Performance
- 100Accessibility
- 100Best Practices
- 100SEO
Have A Product
In Mind?
Tell Us What You’re Building. We’ll Reply Within One Business Day With A Scoped Plan.