-
Notifications
You must be signed in to change notification settings - Fork 0
How AI Was Used
Felines was made by one person. AI was a tool. Every decision, every idea, every ethical choice was human.
Hack the Kitty's rules require "meaningful human direction." This page documents what that means in practice — not as a formality, but because it's the truth of what happened.
Felines is not an AI-generated product. It's a product created by one person with a background in UX Design, Business Analysis, and Service Design, who used AI tools to accelerate technical implementation.
The central insight behind the entire product — that no existing platform speaks to someone who isn't a cat person yet — came from a personal observation. The founder herself didn't know what a cat colony was before starting the project. She was her own primary persona, and that guided every product decision that followed.
Reading "world domination" as "conquer those who aren't allies yet" was a human creative and strategic decision. No instruction, prompt, or tool arrives at that interpretation on its own — it was a deliberate positioning choice. See Product Vision and the README's hackathon-theme section.
Every data point behind the product and this documentation comes from real, verified sources — see Research and Benchmark for the full list with citations: WHO/WSPA, Instituto Pet Brasil, IBGE, Alley Cat Allies/AVMA, the Brazilian Ministry of the Environment, Cobasi Cuida, ABADI, Censo Condominial, and direct legislation (Lei 9.605/98, Lei Sansão 14.064/2020, Lei 15.046/2024). No data point was fabricated, estimated, or generated by AI. Every number has a source, and every source is linked.
The 14 real animal-protection contacts in Natal/RN were researched and verified manually — phone numbers checked, addresses confirmed, organizations identified by their correct name.
Every decision that defines Felines was made by the founder, including:
What to build. The MVP definition, which features made the cut, what was deferred to the Roadmap, what's essential versus nice-to-have. Every feature that exists in the product passed through a conscious human approval — see Design Decisions for 18 of these decisions documented individually, with what was chosen, what was discarded, and why.
What to discard. Several features were considered and rejected. A public shareable colony URL — discarded over animal safety risk. A caretaker leaderboard — discarded because it would create competition where the product needs collaboration. Sharing exact locations — discarded because real cases show cats being harmed after locations were shared online.
What emerged along the way. The floating cat assistant was an idea that surfaced mid-development — inspired by Microsoft Office's Clippy, with the vision of using an AI-generated animated character contextually and non-intrusively (see Behind the Scenes). The decision to use fun curiosities instead of statistics in the assistant was a specific editorial choice, made after noticing serious data already existed elsewhere in the product.
Progressive location blur, framed as physical animal safety rather than just data privacy, was a human formulation — no tool arrives at that framing on its own. See Security for the full technical implementation of that human decision.
The decision not to report people, only situations — avoiding digital vigilantism between neighbors — was a conscious ethical choice (see Design Decisions — D05).
How features evolved. The product after weeks of development is meaningfully different from what was originally planned. Features were added that weren't in the original scope. Planned features were abandoned. The 3-confirmation report-resolution logic, the auto-calculated health index, the anonymous interest system in the resource exchange, and the 3-false-pin-flag ban system all emerged during the process as responses to real problems that surfaced along the way — not from an upfront spec.
The product's colors were chosen from the logo cat — the
connection between the palette (#B66119 and its derivatives)
and the character was a human design decision. See Visual Identity
for the full reasoning, including the discarded name Gatopolis
and why Felines won.
The educational content was reviewed and approved in full. Every article was read, every assistant curiosity was evaluated, translations were checked. The 36-term bilingual glossary was curated by hand.
The decision to use progressive location blur to protect the animals physically — not just user data — was a human ethical stance, informed by real cases of cats being poisoned after their location was shared online. The decision to never return exact coordinates in any public API response, even though RLS already protects the database, was an intentional extra layer of defense, thought through and implemented deliberately — not a default a tool would reach for on its own. Full detail: Security.
Claude Code was used to write the application's code. This included:
- Structuring the 21 routes in Next.js
- Implementing the 83 Supabase migrations
- Building React components from the founder's design decisions
- Initial drafts of the 19 educational articles
- PT/EN translation of the content
What this doesn't include: the founder reviewed code throughout the process — catching inconsistencies, questioning implementation decisions, reverting approaches that didn't reflect the product's intent. Code was never accepted blindly. Every session had active approval, correction, and direction.
Used as a reasoning partner throughout the project — to structure ideas, build prompts, organize documentation. It functioned as a writing collaborator, not a decision-maker.
The cat-assistant video was generated by AI video tools from a detailed prompt written by the founder, with specific art direction: watercolor style, transparent background, a defined movement sequence (walk, lick a paw, blink, turn and leave). The character is Felines' own logo cat. The creative direction was human; the generation was technical.
Projects built with AI raise legitimate questions about authorship, originality, and integrity. Felines addresses these directly:
Real data, not generated. No statistic was invented or generated by AI. Every number has a source, every source has a link — see Research and Benchmark.
Decisions with real consequences. Felines' security choices have physical impact — the location of stray cats can put real animals at risk if handled carelessly. That weight was taken seriously in every product decision, not treated as an abstract data-privacy checkbox.
Reviewed content. Every AI-assisted educational article was read and approved before publishing. No content went live without human review.
Audited code. Security audits identified real vulnerabilities — path traversal and SSRF, both from Aikido Security, plus several found in internal review — that were fixed with an understanding of the problem, not just automatic patch application. See Security and Testing for the full record.
| What was human | What was a tool |
|---|---|
| The idea of speaking to non-cat-people | Code generated by Claude Code |
| The research into data and sources | Initial PT/EN translation |
| Every product decision | Initial drafts of the articles |
| The security architecture | Cat-assistant character video |
| Progressive blur as physical animal protection | React component structure |
| The visual identity and tone of voice | — |
| Curation of the 14 real contacts | — |
| Code review and feature approval | — |
| The new features that emerged during development | — |
| The reading of the "world domination" theme | — |
| Every ethical decision in the product | — |
AI wrote the code. The product is the founder's thinking.