Sources
Domain adapters connect to ephemerides, exchange quotes and arbitrary series.
System design
FENKI architecture
A four-layer pipeline from raw sources to LLM-ready context. Every layer is isolated and every exporter is replaceable, so the core remains stable as new domains are added.
Layers
Four context layers
Data moves through progressively richer representations, from a raw stream to a meaningful prompt.
Exporters
Domain-specific extraction logic understands each source and its semantics.
Transform Kit
Normalization, interval tokenization and markers use reproducible operations.
Context Pack
Prompt blocks, JSON schemas or structured objects are ready for a model.
Principles
Core principles
Four rules keep every pipeline predictable as the kit expands.
Determinism
One input produces one context with no hidden randomness between stages.
Semantic integrity
The meaning of data survives from source to prompt.
Extensibility
A new domain means a new exporter, not a rewritten architecture.
Composability
Transforms assemble into explicit, testable chains.
Source→
Export→
Parse→
Normalize→
Context→
Prompt→
LLM
The FENKI core owns Export through Context. Sources and models stay outside the boundary, keeping the pipeline portable.