Case Studies
Real projects. Real results. Here's what systematic AI evaluation looks like in practice.
AI Evaluation
~3 months
Tencent (via Turing)
Delivering evaluation tasks that shipped into Tencent's published CL-bench (arXiv:2602.03587)
Tasks shipped in published CL-bench paper
Outcome
Multiple frontier LLMs
Models Tested
Publication-grade discriminative tasks
Task Quality
Agent Benchmarking
~2 months
SAP (via Turing)
Building a comprehensive evaluation harness for multi-step tool-use in enterprise AI agents
43 across 6 categories
Tools Evaluated
78 across 5 complexity tiers
Scenarios
4 automated metrics
Scoring Dimensions
MCP-based, CI-integrable
Architecture
AI Evaluation
Ongoing
Independent
Building an automated platform to systematically find failure modes across frontier LLMs
128K calls/day
API Throughput
100K+/day
Automated Evaluations
0.89 Spearman vs expert labels
Judge Correlation
9 systematic categories
Failure Categories