2. MaxBrin,Co-Founder&CEO
CTO Cyber @ Amdocs
Co-Founder & CEO @ Cards
Research & Innovation Lead @ CyberArk
Co-Founder & CEO of Hoger
Chief Architect @ Enabley
About me
3. What is Floom trying to solve?
The gap between AI Potential
and AI Integration
4. Where’s the gap?
No robust dev/dep infra
Technical barrier
No Centralized Management
Inspiration Deficit
5. Existing AI Integration Flow
AI
Model
App
Code
Library
Gateway
Logging, Caching,
Routing, Security etc.
Logic Coating
Data Ingestion, RAG,
Embeddings
Centralization
7. AI Functions
Direct Inference + Agents
Extract Physical Addresses
Generate Speech from Text
Detect Objects in Image
Detect Emotion in Text
Classify PG rating in Video Ask questions about data
Invoke Tools for action Generate Image
12. Why did we choose Milvus DB?
Lowest latency
Friendly, well documented APIs
Predictable under load!
Passed our rigorous production testing
13. How does Milvus DB enable Floom?
Floom relies on Milvus for RAG and general embeddings search (CS/L2/IP)
Default Vector DB for Floom Docker (part of docker-compose)
Default Vector DB for Floom Cloud (Zilliz Cloud)
Full RAG + Multimodal (Video!) experimentation
14. Most companies keep GenAI as POC
Most product managers focus on the classic GenAI examples
(text+code generation)
Generalizing GenAI is hard, but feasible and well worth it.
GenAI integration is currently reserved to GenAI enthusiasts, Data
Science or Innovation team
Using current dev/dep methodologies, GenAI is not yet
production-grade
Key Lessons
15. Roadmap:
AI Definition Markup Language
● Generalized approach towards developing and maintaining AI integrations
● Distilled to “AI Functions” (including both agents and direct inference)
● Easy to understand, build, debug and maintain across versatile teams
● Designed for developers with no to little prior AI/GenAI knowledge
● Easily packed in a distributable format
○ Definition (AI Functions: Inference + Agents)
○ Configuration
○ Orchestration