Concept
DODIL is an AI cloud designed to run data-heavy and GPU-intensive workloads reliably and at predictable cost.
Instead of stitching together many unrelated services, we focus on a small set of platform layers that matter most for AI products:
- Infrastructure foundation (compute, storage, networking, orchestration)
- Cloud APIs (clean APIs + SDKs for developers)
- AI-native services (the Vector Stack today, more coming next)
What DODIL optimizes for
AI workloads tend to be:
- Data-heavy: large files, streaming pipelines, multimodal datasets
- Compute-intensive: GPUs for embedding, inference, and training
- Latency-sensitive: interactive apps, retrieval, and agent loops
- Cost-sensitive: storage, bandwidth, and token-based usage add up quickly
DODIL is built to keep performance consistent while making cost transparent.
The platform layers that matter
1) Infrastructure foundation
This layer is responsible for running workloads safely and predictably across regions and tenants.
It includes:
- Compute & GPU scheduling for AI jobs and services
- Storage for datasets and artifacts
- Networking optimized for service-to-service traffic
- Isolation & tenancy for secure multi-tenant operation
- Identity and policy enforcement as a default, not an add-on
- Observability and metering so usage and performance are measurable
2) Cloud APIs
DODIL exposes developer-friendly APIs and SDKs to:
- provision and manage resources
- run jobs and pipelines
- store and query data
- deploy and operate AI workloads
These interfaces are designed to stay stable as the platform evolves.
3) AI-native services
DODIL is building higher-level services on top of the foundation. The first release focus is the Vector Stack:
- VNG (Vector Ingest): ingest, chunk, enrich, and embed multimodal data
- VBase (Vector Database): store vectors + metadata with fast vector and hybrid search
Coming next:
- VI (Vector Intelligence): drift, anomalies, quality signals, and performance insights
- MCP (Model Control Plane): deploy, route, observe, and govern model workloads
What this enables
When these layers work together, vertically integrated, you get:
- Lower cost at scale through better utilization and fewer unnecessary layers
- Predictable performance on infrastructure tuned for AI workloads
- Faster iteration with consistent primitives across ingestion, storage, and serving
- Operational clarity with metering, observability, and policy built in