LogicVisor
An AI-powered code review platform giving instant algorithmic feedback without requiring signup.
Built with abuse-prevention as a first-class concern — AST-based duplicate detection keeps free-tier costs sustainable.
[DIR]: /metadata
[DIR]: /problem_statement
Getting quick, structured feedback on algorithm implementations usually means paying for a full AI subscription or waiting on a human reviewer. LogicVisor gives anonymous users a free, rate-limited trial with real model feedback, with paid tiers for people who want more.
The core technical challenge is abuse prevention without login friction — anonymous access is inherently exploitable, so cost control has to happen at the infrastructure layer rather than through account gating.
[DIR]: /architecture
Built on Next.js 15 and React 19 with Supabase (Postgres) for persistence. The review pipeline integrates Google Gemini and Groq's Llama models for feedback generation. Session-based anonymous access is tracked with browser fingerprinting; submitted code is parsed with Tree-sitter to build an AST, then hashed with SHA-256 so duplicate or near-duplicate submissions get served cached results instead of re-hitting the model APIs. An admin dashboard tracks model performance, conversion, and usage patterns across the three-tier access model (anonymous, free signed-in, Pro).
[DIR]: /key_decisions
- Session-based anonymous trials over forced signup: Removing the signup wall increases top-of-funnel usage, but shifts the abuse-prevention problem onto infrastructure — fingerprinting plus rate limiting replaces account-based gating.
- AST fingerprinting for duplicate detection: Tree-sitter parsing plus SHA-256 hashing catches near-duplicate submissions (renamed variables, reformatted whitespace) that a raw string hash would miss, so cached results get served instead of paying for a redundant model call.
- Multi-model integration (Gemini + Groq Llama): Running two model providers gives redundancy if one has an outage or gets rate-limited, and lets different access tiers map to different cost/quality tradeoffs.
[DIR]: /outcomes
- Live and handling real anonymous and signed-in traffic across three access tiers.
- Admin dashboard tracking model performance, conversion metrics, and usage analytics.
- Abuse-prevention pipeline (AST fingerprinting + rate limiting) keeping free-tier costs sustainable.