Search
Enter a vertical, competitor, or keyword, or provide a manual ad media URL for the analysis run.
Featured work · AI product case study
AngleScope turns ad-library research into a creative intelligence workflow.
I built AngleScope for affiliate and performance marketing teams that need to find winning ad angles, understand why they work, and turn those patterns into new testable creative concepts without pretending to have private ad-account data.
The Render demo is hosted on a free instance, so first load can take a moment if the service is asleep.
What it demonstrates
Marketing judgment, AI implementation, and data realism in one product.
AngleScope is deliberately not a fake spend dashboard. As an outside builder, I did not have private Meta, Google, TikTok, Taboola, landing-page, or lead-quality data. So I focused on the part of the buyer workflow that can work honestly today: public ad examples, user-supplied creative, structured AI deconstruction, and evidence-backed concept generation.
AIOpenAI-backed ad deconstruction and creative generation when configured.
ZodStructured schemas and deterministic fallbacks keep output usable.
LiveRender deployment runs the full server-side Next.js workflow.
Problem
Media buyers already do this manually.
Creative angle discovery is often one of the highest-leverage parts of affiliate media buying. Buyers scroll ad libraries, save examples, look for ads that appear to keep running, infer which hooks and proof patterns are working, and translate those patterns into the next batch of tests.
That work is valuable, but it is slow, uneven, and easy to lose in screenshots and scattered notes.
Product thesis
AngleScope turns that research loop into software: collect examples, classify what is happening, cluster recurring angles, rank them with evidence, and generate offer-specific concepts that a buyer can actually test.
The goal is not to replace the operator. The goal is to give the operator a sharper starting point.
Workflow
From public examples to creative concepts.
Ingest
Load validated seed examples and best-effort public examples from TikTok Creative Center when the server app is running.
Deconstruct
Classify hooks, emotional angles, formats, offer mechanics, CTAs, claims, and compliance risk.
Cluster
Group recurring winning angles and rank them with evidence from source ads and strength scoring.
Generate
Turn selected angles and offer details into new creative concepts, copy, briefs, and image direction.
Export
Make the output usable outside the demo with JSON and CSV export paths.
Architecture
Built as a real product surface, not a one-off prompt demo.
Next.js App Router + TypeScript
Tailwind UI workbench
Zod schemas for API and model output
Source adapters for seed, manual, and TikTok data
Prisma + Postgres persistence scaffold
Build decisions that matter
- Kept private spend, ROAS, and lead-quality claims out of the demo because those require internal credentials.
- Used source adapters so public, seeded, manual, and future private data can follow the same analysis contract.
- Constrained AI behind structured schemas and deterministic fallbacks so the workflow remains demoable when APIs fail.
- Designed the UI around the operator loop: search, inspect, choose an angle, generate concepts, and export.
Current live build bar
- Demo
- Full server-side Next.js app deployed on Render.
- Sources
- Seed examples, manual ad URL input, and best-effort TikTok Creative Center ingestion.
- AI path
- OpenAI-backed visual and metadata deconstruction plus structured generation when configured.
- Fallbacks
- Deterministic, evidence-grounded concepts and a static GitHub Pages fallback.
- Output
- Ranked angle clusters, creative concepts, briefs, copy, image direction, JSON, and CSV.
What I would build next
Close the loop between creative intelligence and business outcomes.
The next version would connect internal performance data so AngleScope can learn from actual winners and losers, not only public examples. That means joining ad creative attributes to ROAS, CPL, lead quality, approval risk, and funnel drop-off.
From there, it becomes a creative operating system: discover public winners, deconstruct internal winners, generate compliant variants, match each angle to landing-page or advertorial treatments, and recommend the next tests based on observed performance.
Near-term roadmap
- Persist analysis runs and generated concepts in Postgres.
- Harden the TikTok adapter with cached snapshots and monitoring.
- Add Meta Ad Library as the next source adapter.
- Upgrade manual URL input to stored uploads.
- Add compliance pre-flight checks for risky claims.
- Create weekly angle-opportunity reports for the buying team.
Portfolio fit
Why this belongs next to RevOps and AI GTM systems work.
AngleScope is useful as featured work because it shows the same operating pattern behind strong GTM systems: understand the real workflow, define the data boundaries, build a repeatable system, and keep the output practical enough for an operator to use.