Source input
Reviewers can enter a competitor URL, paste source text, or run a saved sample without an external service.
Flagship AI productFive-day independent build
Aegis analyzes competitor landing pages, drafts five campaign concepts, and reviews the copy for risk in a separate step.
I built Aegis to handle malformed model output, failed extraction, and reviewer questions about source and risk. The workflow around the model matters as much as the generated copy.
Next.jsTypeScriptGeminiFirecrawlZodRisk review
Why I built it
Creative teams need a faster first pass and a clear review path
Competitor pages contain useful information, but teams still need time to review positioning, claims, hooks, audience assumptions, and conversion logic.
Aegis applies the same steps to each source and keeps concepts, risk flags, and errors visible for review.
Product walkthrough
Review each stage of the workflow
The interface separates source input, analysis, concepts, and risk review instead of hiding them in one response.
Structured strategy
Spyglass returns a consistent analysis of the offer, audience, hooks, claims, objections, and creative opportunities.
Visible risk review
Shield reviews generated copy separately and explains which phrases need closer attention.
Lower-risk alternatives
Suggested rewrites keep the concept while the reviewer accepts, revises, or rejects the wording.
Key design decisions
The controls are part of the product
Give each AI stage a contract
Aegis separates strategy, concept generation, and risk review. Zod validates each response before the interface uses it.
Treat failure as a product state
Saved samples, manual input, stage-specific errors, and preserved results keep the workflow usable when extraction or a model stage fails.
Keep human review visible
The interface shows source context, pipeline history, flagged language, and lower-risk options. Shield supports review; it does not promise legal or platform approval.
Product pipeline
Four stages with clear handoffs
- 01
Extract
Collect page content from a URL or manual source text.
- 02
Analyze
Identify the offer, audience, positioning, hooks, and claims.
- 03
Generate
Draft five campaign concepts from the validated analysis.
- 04
Review
Flag policy risks and suggest lower-risk wording.
Source traceability needs more work
The prototype uses the supplied source, but it does not yet link every finding or generated claim to a specific passage.
Link review actions to the evidence
The next version would add source-linked excerpts, saved projects, side-by-side comparisons, reusable brand rules, and clear accept, revise, or reject actions.
Aegis shows my AI product approach: define the task, validate the output, plan for failure, and keep review visible.