Technical staff gaining approved access to AI
Previously had effectively no approved access
EVIDENCE LIBRARY
Knowledge for deployable enterprise AI
Patty publishes more than product explanations. We document what changed in operation, which technical choices mattered, what has been verified, and what remains outside the evidence.
01CUSTOMEROperational change
02METHODValidation method
03SYSTEMTechnical design
04LIMITSScope and caveats
Patty perspective
Enterprise AI is an operating-system problem, not a model-selection exercise
Connecting a capable model does not resolve enterprise data, authority, execution, or accountability. When AI participates in real work, the organization must be able to explain who requested an action, what information it accessed, what it executed, and who approved it.
This hub brings together customer experience, technical measurement, security doctrine, and open specifications for designing that operating system. We keep sales claims separate from evidence used to evaluate a system.
Evidence system
What we publish as evidence
Different questions require different forms of evidence.
01
Customer stories
Anonymized records of prior conditions, scope, implementation, observed change, and difficulty.
02Technical reports
Benchmarks and measurements with hardware, build, data, method, and explicit caveats.
03Security & governance
How AI identity, data boundaries, approval, execution, and evidence connect at runtime.
04Open technology
Specifications and conformance paths that let third parties inspect delegation and execution receipts.
05Blog
Current perspectives from product, technology, security, and enterprise AI operations.
Reader paths
Read by responsibility
The first question changes with the reader’s accountability.
- 01Customer evidence → AI workforce → governance →
Executives
Does AI investment translate into operating outcomes and accountable ownership?
- 02Security & governance → open specifications →
Security & compliance
How are data, models, tool execution, and approvals controlled and proven?
- 03Mirr Code → monographs → measurement reports →
Engineering
How do agents work in Korean development environments and legacy systems?
- 04Models & training → inference → benchmarks →
AI & infrastructure
How should training and GPU inference operate under real constraints?
Featured customer evidence
From prohibited AI to a controlled everyday capability
A public-sector environment that could not use external generative AI brought model serving, the coding agent, policy, and execution evidence inside.
Read the public-sector story →- Technical personnel
- Approximately 70
- Deployment
- On-premise · controlled network
- Inference
- Internal distributed GPU infrastructure
- Expanded operation
- Approximately 15–20 specialized agents
Throughput for selected AI-suitable workflows
Not organization-wide productivity; compares manual baselines for code understanding, tests, and repetitive changes
First useful understanding of unfamiliar code
Selected cases that previously consumed most of a day
Research & reports
The method comes before the number
A result matters only when the reader can see what was measured, where, how, and what the result cannot claim.
Finished-path measurements for an engine-native agent browser
A measurement of the complete observe→classify→park→approve→dispatch path when an AI agent is embedded in the browser engine. Semantic snapshots remained around 14–16ms in the tested fixtures, and the governed engine cycle completed in under 25ms.
Read report →Designing a benchmark corpus for coding agents
Design notes for an end-to-end corpus stratified across eight workload classes, a 50-instance SWE-bench Verified subset, and a context-retention A/B harness, including anti-guess tasks and scorer validation.
Read report →BLOG
All Blog postsLatest from the Blog
From Matching Models to Recruiting Agents: AI Hiring Systems, Evaluation, and Governance
Tracing how AI hiring moved from scoring profile pairs to multi-stage recruiting workflows, and what a systematized review of 40 works says about evaluation evidence and governance.
EXAONE Forecast for Finance: A Linear Foundation Model Designed Apart for Finance
Tracing how EXAONE Finance pairs an attention-free linear backbone with causal convolution, group-aware pooling MLP, and masked context augmentation to cover long many-channel panels and rank first on all three FinVerse tiers.
Tau Tau Bench: A Benchmark That Scores the Whole Job of Building Agents
A new evaluation setting that turns building a customer service agent into the task for coding agents. A close read of its realistic starting conditions, held-out user scoring, low pass rates, and what they mean.
Publication standard
Five questions attached to every claim
Who says it, what was observed, where it applies, how it can be reproduced, and what we are not claiming.
- 01
Source
Separate customer report, pilot observation, internal measurement, and public research.
- 02
Conditions
Record period, participants, hardware, data, and build.
- 03
Method
Explain the procedure that produced the measurement or judgment.
- 04
Result
Separate observed change from its implications for a decision.
- 05
Limits
State where evidence does not support organization-wide, causal, or certification claims.
Explore by problem
Start with the problem your organization is trying to solve
On-premise and closed-network AI
Keep source, prompts, and inference inside the controlled environment.
Korean AI coding
Execution designed around HWP/HWPX, IME, initial consonants, Korean instructions, and local engineering practice.
AI security and governance
Identity, policy, approval, audit, and verifiable execution evidence.
AI agent management
Operate responsibility, tasks, escalation, and resources across many agents.
GPU serving and inference
Distributed serving, scheduling, KV cache, and MoE execution architectures.
Korean RAG and document understanding
Retrieve, extract, and validate Korean documents and business context.
NEXT STEP
Discuss an evaluation Begin with your operating constraints
Share your industry, data boundary, network, development environment, and approval structure. We will define the evaluation scope and success criteria before the product demonstration.