Published · Last updated · By Pedro Shin, Founder
AlgoMaxxing is an AI-powered social media profile analyzer operated by Pedro Shin from São Paulo, Brazil. This document explains, in detail, what technology powers the product, which public data signals we ingest, how we score profiles on the S to D tier scale, and how we handle your data under LGPD and GDPR. No marketing claims appear here — only what actually runs in production.
AlgoMaxxing is built on a small, deliberately boring technology stack. The user-facing web application runs on Next.js 16 with React Server Components, deployed to Vercel's edge network for global latency under 200 milliseconds. The AI kernel is a FastAPI service written in Python 3.12 and orchestrated with LangGraph for agent state transitions. Content analysis is performed by frontier large language models from Anthropic (Claude families) and OpenAI (GPT-4 family), with model routing chosen per task type. Specialized vision models assess profile grids, thumbnail aesthetics, and bio layouts — the same Claude and GPT vision endpoints used for the text pipeline. Social profile ingestion flows through Bright Data for official public API access and managed HTML fallbacks, with per-platform rate-limit awareness. Semantic retrieval across our internal benchmark corpus is handled by Postgres with the pgvector extension and HNSW cosine indexing. Stripe handles all billing and invoice flows. Supabase provides Postgres hosting, email authentication, and row-level security. Logging and performance telemetry run through Vercel Analytics and Sentry.
AlgoMaxxing reads public profile data from six social platforms: Instagram, TikTok, YouTube, X (formerly Twitter), LinkedIn, and Facebook. For each platform we ingest a fixed set of public signals. First, the recent post count over the trailing thirty days. Second, the engagement rate, computed as likes plus comments divided by total followers, averaged across the most recent public posts. Third, the posting cadence, measured as the median interval between consecutive public posts. Fourth, the follower growth delta over the trailing thirty days, inferred from deterministic snapshots. Fifth, the content-format mix — the ratio of video to image to text-only posts. Sixth, the bio and avatar for vision analysis. We use the platform's official API whenever available, falling back to authorized paid data partners (for example Bright Data) and then to public HTML scraping for resilience. Private accounts, direct messages, and any non-public content are strictly out of scope and are never ingested. We never store, request, or accept your social platform passwords or OAuth refresh tokens.
Every scanned profile is mapped to a letter tier using a weighted blend of three signal families inside the profile's detected niche. Engagement — likes plus comments divided by followers, averaged across recent posts — counts for forty percent of the weighted score. Posting consistency — a measure of regular cadence that rewards steady output over the past thirty days and penalizes week-long gaps — counts for thirty percent. Audience-growth velocity — the delta in followers over the trailing thirty days — counts for the remaining thirty percent. The weighted score is then mapped to a tier: S is the top ten percent within the niche, A is the seventy-fifth to ninetieth percentile, B is the fiftieth to seventy-fifth, C is the twenty-fifth to fiftieth, and D is the bottom quartile. Niches are detected automatically from content signals (captions, hashtags, on-image text, video transcripts), so your tier is always computed against peers with similar audience expectations. Scores are comparative within niche, not absolute, and recalibrate as the peer corpus grows.
AlgoMaxxing is operated from Brazil and complies with Lei 13.709/2018 (LGPD) as its primary privacy regime, and with GDPR for European users. Scan results are cached for a maximum of twenty-four hours in hot storage to accelerate re-scans of the same handle, then aged out of the active data store. No user-submitted data is used to train third-party AI models — we explicitly opt out on the vendor side where such controls are exposed. Payment data is processed entirely by Stripe and never persists on our servers. Under LGPD Articles 9, 10, and 11, you have the right to access, correct, port, anonymize, and delete personal data we hold. To file a request under LGPD or GDPR, email pedromaschio.shin@gmail.com with your user email or handle; we respond within the statutory fifteen-day window. We do not sell personal data, and we do not share it with third parties beyond the sub-processors listed in the Technology stack section above.
AlgoMaxxing scores are informational, comparative, and non-authoritative. They reflect our best current estimate of a profile's relative standing within its detected niche, computed from public signals and AI interpretation of visual and textual evidence. AI models make mistakes — they can misread niches, miscount engagement on ambiguous content, or underweight emerging formats. Scores should inform your content decisions, not dictate them; creators who treat a single tier card as a verdict miss the broader picture. We make no guarantee of audience growth, revenue, brand deals, or algorithmic reach. The scoring rubric is subject to revision as we learn more about platform algorithms and as the benchmark corpus expands. A score generated today may differ from a score on the same profile generated next quarter, even with identical content, because the peer distribution shifts. If you spot a clearly wrong score, email pedromaschio.shin@gmail.com with the scan ID and we will investigate. This page reflects the state of the product as of the last documentation update listed in our git history.