From PDF to Insights: How GearCheck Analyzes Your Blood Work

Digital PDFs are parsed directly using text extraction — marker names, values, and reference ranges are identified programmatically. No OCR needed, no quality loss. This is the fastest path through the pipeline.

Scanned PDFs and images — roughly one in three uploads — go through OCR (optical character recognition) to convert the image into machine-readable text. The system detects when OCR is needed automatically and adjusts its parsing strategy.

A preflight check counts how many markers were extracted. If the count is too low — for example, if you uploaded an image of a blood pressure log instead of a lab report — you are alerted before the analysis starts. This saves you from waiting five minutes only to get an error.

Within seconds, every marker in your report is identified and structured into a standardized format, regardless of which lab generated the original PDF.

This is where the heavy lifting begins. Extracted text is passed through two parallel engines that work together to ensure no marker is missed:

Regex extraction — pattern-matching pulls out common markers with known naming conventions. This runs first and captures the low-hanging fruit: well-labeled panels from standard lab formats like "Comprehensive Metabolic Panel" or "Lipid Panel." It is fast, deterministic, and catches the most common markers instantly.

Gemini AI extraction — a large language model reads the entire document in context. It handles unusual panel names, non-standard layouts, foreign language reports (Dutch, German, French, Spanish — all common in European labs), and markers that do not match any pre-defined pattern. This catches everything the regex pass misses.

Both passes are cross-referenced. Markers found by both engines are flagged as high-confidence. Markers found by only one pass are validated against known marker lists and typical units. The result is a structured dataset of every marker in your blood work, mapped to canonical names and values — regardless of how the original lab labeled them.

Before anything else, the system verifies whether your blood work is analyzable. This quality gate prevents analysis from proceeding on incomplete or implausible data:

Minimum marker threshold — a complete profile requires at least some markers from each key system: kidney, liver, cardiovascular, glucose metabolism. If too many systems are missing markers, the panel is flagged as incomplete.

Critical markers present — if markers like creatinine, AST, ALT, HDL, and hematocrit are missing, the system flags the panel as insufficient. These five markers are the minimum viable set for any meaningful AAS health assessment.

Value plausibility — extreme outliers that suggest transcription errors are flagged for review (e.g., a hematocrit of 85% is almost certainly a data error, not actual blood thickness). This catches OCR errors and lab typos before they reach your report.

Each extracted marker is now compared against GearCheck's reference database — but not the standard lab ranges your doctor uses. The system applies athletic reference ranges, meaning the comparison starts from a realistic baseline for someone who trains, eats a high-protein diet, and may use AAS:

Marker status is determined using a four-tier system: ATTENTION (mildly out of athletic range, worth watching), ACTION (significantly out of range, needs investigation), MONITOR (trending in the wrong direction over time), or REVIEW (affected by known confounders like muscle mass, training, or AAS pharmacology, requiring additional context to interpret).

Trends from previous uploads are incorporated — a marker that has been stable for 6 months gets a different interpretation than one that spiked suddenly. This is where having multiple uploads pays off: the system can distinguish your chronic state from acute changes.

This is the step that makes GearCheck fundamentally different from every other blood work analyzer. The system applies 13+ contextual rules tailored to athletes using AAS. These rules prevent the false positives that athletes routinely get from standard medical interpretation:

The muscle leak rule: elevated AST/ALT with elevated CK and normal GGT = muscle origin, not liver damage. This is the single most commonly misdiagnosed pattern in our database.

The kidney adjustment rule: elevated creatinine with normal Cystatin C in high-muscle-mass individuals = artifact, not kidney impairment. This rule alone saves dozens of unnecessary nephrology referrals per week.

The AAS lipid rule: suppressed HDL under AAS is a pharmacological effect — the system shifts focus to ApoB and LDL particle count, which are better predictors of actual cardiovascular risk.

The hematocrit rule: hematocrit is interpreted with compound risk, blood pressure, and the athlete's baseline over time, not against a single population-based cutoff.

Finally, everything comes together into a complete health analysis:

A Health Score is calculated — a single number (1-5) that summarizes overall health status across all body systems, weighted by clinical significance.

The report is assembled — organized by body system: kidney, liver, cardiovascular, hematology, hormones, metabolic, and electrolytes. Each section gives you the key markers, their values, their status, and what to do about them.

Narration is generated — each section includes a written analysis that explains what each marker means in your context, not just the number. The narration connects related markers, highlights patterns, and explains why certain values are expected given your protocol.

Key findings are highlighted — the most important patterns and changes across all systems, prioritized by clinical relevance.

The result is a comprehensive health analysis that would take a human specialist 45-60 minutes to produce, delivered in under 5 minutes.