Build, test, evaluate, and improve AI applications.
Practical engineering tools and reproducible research for AI agents, RAG pipelines, prompt evaluation, and MCP security.
Featured Engineering Tools
Core diagnostic and evaluation utilities for modern AI stacks.
MCP Security Scanner
Audit Model Context Protocol configuration files for shell injections, unpinned packages, and excessive filesystem permissions.
Prompt Injection Scanner
Scan user prompts and untrusted RAG chunks for direct instruction overrides, jailbreak personas, and markdown exfiltration.
RAG Retrieval Debugger
Analyze token overlap and sentence-level attribution between retrieved document passages and generated answers to catch hallucinations.
RAG Chunking Playground
Experiment with document splitting strategies, target token chunk sizes, and boundary overlaps for vector databases.
LLM Response Diff
Compare baseline model outputs against candidate prompts to detect token drift, formatting changes, and semantic regressions.
Agent Loop Detector
Detect infinite execution loops, cyclical tool calls, and runaway retry patterns in multi-step autonomous agent traces.
AI API Cost Calculator
Calculate token pricing, prompt caching discounts, and monthly operational costs across OpenAI, Anthropic, and Gemini models.
Context Window Calculator
Model token consumption and attention saturation across system prompts, retrieved passages, conversation history, and completion limits.
Engineering Tools Directory
All utilities are grouped into structured technical disciplines to help you diagnose and optimize every layer of your AI stack.
Empirical Research & Failure Modes
Reproducible test case specifications for production AI failure modes.
AI Failure Test Library
A catalog of structured test specifications designed to stress-test AI applications before deployment. Includes test procedures for unbounded agent loops, multi-hop context compression loss, schema hallucinations, and prompt injection vectors.
Engineering Guides & Architecture
In-depth technical writeups on tokenization, latent geometry, and conditioning math.
Text-to-Image Prompt Architecture
How cross-attention layers process token order, why comma-separated tag soup fails, and how to structure descriptive clauses.
Read guide →Latent Aspect Ratios & Training Buckets
Understanding SDXL and Flux bucket resolutions, divisible-by-64 geometry, and preventing multiple torso/limb hallucinations.
Read guide →CFG Mechanics & Negative Guidance
Examining Classifier-Free Guidance vector math, when negative prompting strips desirable detail, and modern alternatives.
Read guide →Start Testing Your AI Systems
Deterministic utilities, zero account sign-up, and 100% free access. Browser-based tools are designed to process supported inputs locally in your browser, minimizing application-controlled transmission of tool content.