🖥️
Best Viewed on Desktop
The technical documentation and app previews are optimized for desktop viewing.
← Back to Home🔍
System Overview
Zentinel-audit v4.3 — AI-augmented smart contract security integrating 28 specialized tools through the Semantic Refinement Pattern. "Tools detect, AI understands."
28
Security Tools
10 trad + 6 LLM + 12 drain
669
Validated Findings
53 DeFi contracts
7.0×
Speedup
162 → 23.1 min
0.86
F1 Score
Prec 0.89 · Rec 0.83
68.4%
GAEV Critical+High
39/57 exploits verified
54%
FP Reduction
57% → 26%
$0.03
Avg LLM Cost
Per contract analyzed
ℹ️ The Semantic Refinement Pattern
A software architecture where traditional analysis tools execute in parallel while LLM agents provide semantic post-processing. The pattern embodies the principle: "Tools detect, AI understands." Traditional tools provide high recall through syntactic pattern detection; LLMs provide high precision through semantic understanding. Together: F1 = 0.86, 54% fewer false positives.
Pattern Properties (Formal Definition — Pattern 2)
P1
Tool Independencen ≥ 2 analysis tools T = {t₁, ..., tₙ} execute independentlyP2
Parallel ExecutionTools execute concurrently: time(T) ≈ maxᵢ(time(tᵢ))P3
Schema NormalizationA function φ : ∪ᵢ Oᵢ → S maps tool outputs to unified schema SP4
Sequential Refinementm ≥ 1 LLM agents A = {a₁, ..., aₘ} process findings sequentiallyP5
Detection SeparationLLM agents never perform primary vulnerability detectionDesign Principles
Separation of Concerns
Detection (tools) is separated from understanding (LLMs). Each component has a single responsibility.
Defense in Depth
Multiple tools with different techniques provide overlapping coverage. No single tool failure compromises the analysis.
Graceful Degradation
If LLM services are unavailable, tool outputs remain valid — just unrefined. The system degrades, not fails.
Composability
New tools added by implementing the output adapter. New agents inserted into the pipeline. Plug-and-play architecture.
7 Contributions
C1Semantic Refinement Pattern — formalized architecture (§3)
C2Pure Dispatcher — zero domain logic orchestration (§4)
C3Unified Finding Schema — cross-tool reasoning (§5)
C46-Agent Pipeline — decomposed complex analysis (§6)
C5Zentinel-audit v4.3 — 28 tools + 12 drain modules (§7)
C6GAEV — 68.4% automated exploit verification (§7.3)
C7Empirical — 7.0× speedup, 54% FP reduction, F1 0.86 (§8)