ChickenTruck

Knowledge Engineering for Python — turning information into structured, traceable knowledge, one nugget at a time.

Status: pre-alpha, architectural stub. The public API below exists so the intended shape of the project is visible and importable, but the real extraction, validation, and storage logic is not implemented yet. Nothing in this package should be relied on for production use.

Purpose

ChickenTruck's job is to take raw information and turn it into structured, validated, traceable knowledge — atomic facts with sources, confidence, and temporal validity attached, ready to feed a graph database, a relational database, a RAG pipeline, or an agent.

The problem

Extracted information is not automatically trustworthy. A model or pipeline that pulls a fact out of a document has no idea, on its own, whether that fact is well-formed, whether it conflicts with something already known, whether it's still true, or where it even came from. ChickenTruck exists to put a disciplined pipeline between "information we found" and "knowledge we're willing to act on."

Knowledge graphs are one important consumer of that output, but ChickenTruck is not a knowledge-graph library — it's a knowledge engineering toolkit. The graph is one possible destination, not the definition of the project.

Food Truck Fleet context

ChickenTruck is a member of a fleet of independent, food-themed Python packages. Each truck owns one stage of a data pipeline and none of them import each other — they compose only through plain data (DataFrames, and here, structured knowledge objects) at the application layer:

• SushiTruck — ingestion (streaming sources, REST APIs, file/object stores)
• ThaiTruck — DataFrame cleaning and transformation
• RamenTruck — modeling / ML
• BentoTruck — agent orchestration
• FishTruck — anomaly / data-quality detection
• ChickenTruck — knowledge engineering (this package)

Planned responsibilities

• Entity extraction and entity resolution
• Relationship / triple extraction
• Atomic fact and claim representation
• Candidate assertions with source, confidence, and temporal metadata
• Ontology / schema validation
• Conflict detection and knowledge reconciliation
• Provenance and source tracking throughout
• A backend-independent store for accepted knowledge
• Export shaped for graph databases, relational databases, RAG systems, and agents

What ChickenTruck does not own

• Raw data acquisition — that's SushiTruck
• General DataFrame cleaning — that's ThaiTruck
• Machine learning / model training — that's RamenTruck
• Agent orchestration or agent memory — that's BentoTruck
• General-purpose anomaly / data-quality detection — that's FishTruck

ChickenTruck is also meant to stay independent of any one LLM provider, graph database, relational database, vector database, NLP framework, or agent framework. Anything backend-specific will eventually live behind an optional adapter, not in core.

Preliminary menu

Names describe what each piece actually does, not just a theme — see PROJECT.md for the full design notes.

Only these four are scaffolded today, and only lightly — see PROJECT.md for what's designed versus what's still open.

Conceptual pipeline

Raw Information
      |
      v
chicken_nuggets      (atomic knowledge)
      |
      v
chicken_tenders      (candidate assertions)
      |
      v
grilled_chicken      (validation)
      |
      v
chicken_stock        (accepted, structured knowledge)

This is a conceptual direction, not a finalized API contract.

Installation

Not yet published. Once available:

pip install chickentruck

Development status

Pre-alpha. The name and package structure are reserved; the knowledge engineering architecture is being designed deliberately before it's built, rather than bolted together module by module. Current behavior:

• chickentruck.nuggets.Nugget — a plain subject/predicate/object dataclass, usable today
• chickentruck.tenders.ChickenTender — a plain candidate-assertion dataclass, usable today
• chickentruck.stock.ChickenStock — a minimal in-memory store, usable today
• chickentruck.nuggets.extract_nuggets() and chickentruck.grilled.grill() — raise NotImplementedError; the extraction and validation engines haven't been designed yet

Design principles

• Don't claim capabilities that don't exist yet — an unimplemented feature raises NotImplementedError rather than faking output.
• Don't duplicate what another fleet member already owns.
• Stay backend- and provider-independent in core; push anything specific (an LLM, a graph database, a vector store) into optional adapters.
• Food names describe function, not just theme.

License

MIT — see LICENSE.