Accelerating innovation using tiered AI Architecture
summary
A portable microservice ML framework called Minerva (and an open mini fork contain) serves model code across legacy and cloud SaaS suites by adding a JSON-configurable data layer and three porting modes: no-code mapping, low-code GPT-assisted tuning, and full high-code rewrites. Experiments porting marketing models across products (Oracle, MailChimp, Constant Contact) show model reuse yields 1.15–2× faster adoption and up to 2× development speed, supports TensorFlow/scikit-learn, runs docker single-node for edge and sustains ~10 requests/sec.
AI-written summary · about
chapters (14)
- 0:00Tiered AI architecture intro
- 1:36Microservice ML model serving
- 3:39Minura and Contain frameworks
- 5:08Differences from ML frameworks
- 7:00Research methodology & datasets
- 8:52Enterprise software experiments
- 10:18Three-layer framework details
- 12:28Microservice layer & data input
- 13:36Code porting layers
- 15:46Innovation experiment results
- 17:10Model code porting improvements
- 18:13Generative AI porting efficiency
- 19:16Performance & real-time use
- 19:36Conclusion
related talks
26mManaging ML Models @ Scale
21mHarnessing Microservices for Scalable, Fault-Tolerant ML Systems
28mAccelerating AI at Scale with Data
38mFantastic ML apps and how to build them
30mEssential Ingredients in Scaling Organizations for ML
41mInfrastructure for ML Applications
51mSpeeding up deployment of machine learning models
12mPreview: ML Serving