Run-Time Prevention of Software Integration Failures of Machine Learning APIs
summary
SmartGear is a run-time adapter that sits between ML cloud APIs and application code, inserted via bytecode instrumentation, that detects hierarchy/perspective/focus mismatches and incorrect outputs by clustering, segment+ensemble, and validation across APIs, inputs, and software. The evaluated prototype detects about 70% of incompatible API outputs and raises application execution correctness from 84% to 95%, preventing roughly two thirds of potential integration failures with low runtime overhead.
AI-written summary · about
chapters (7)
- 0:00ML API integration challenges
- 3:35API integration failure causes
- 8:06Incorrect API outputs and sources
- 10:36SmartG: Run-time adapter
- 14:12SmartG evaluation vs baselines
- 15:58Summary and demo invite
- 16:41Q&A
related talks
22mFormal Runtime Error Detection During Development in the Automotive Industry
50mFrom Prompt Hacking to Architectural Determinism: Engineering Reliable GenAI Systems
26mDetecting and Mapping Code Defects Introduced by Generative AI Using CodeLogic & Neo4j
58mPatterns, Predictions and Programming
30mAI Application Architecture: Building for Scale and Efficiency
23mStop Making LLMs Deterministic: A Spec-Driven Software Factory
17mTraining with Confidence: Catching Silent Errors in Deep Learning Training with...
45mWhere AI Meets Code