Chunking, embeddings, indexing, retrieval, reranking, and RAG evaluation.
AI & ML Engineering Lab · Generative AI & LLM Engineering
RAG Factory
Chunking, embeddings, indexing, retrieval, reranking, and RAG evaluation.
RAGFAMEL Factory
AMEL · Factory Portfolio
A capability in the engineering lifecycle.
This Factory is a specialist tool of its Lab and provides a bounded technical capability. An Engineering Contract defines the need; Foundations and Baselines determine the model. Execution and outcomes remain versioned and verifiable.
Individual Factory profile
RAG Factory as a capability in the AI & ML Engineering Lab.
The RAG Factory (RAGF) groups the documented responsibility area “Chunking, embeddings, indexing, retrieval, reranking, and RAG evaluation” within the AI & ML Engineering Lab. It belongs to the “Generative AI & LLM Engineering” domain and is described as an independently versioned Factory capability within that scope.
This profile describes the documented target model of this specific Factory. It claims neither a production implementation nor functions that are absent from the Factory record.
- Chunking
- embeddings
- indexing
- retrieval
- reranking
- RAG evaluation
- Domain
- Generative AI & LLM Engineering
- Engineering Lab
- AI & ML Engineering Lab
- Roadmap position
- Phase 3 - Generative AI, priority P2
- Documented status
- Concept in the Orthevs Factory catalog
Mission and responsibility
What the RAG Factory is intended to cover.
The following responsibility areas are derived from the versioned profile of the RAG Factory. They make the subject “Chunking, embeddings, indexing, retrieval, reranking, and RAG evaluation” concrete for this Factory and separate it from general platform work.
- Responsibility area · 01
Documented focus: Chunking.
- Responsibility area · 02
Documented focus: embeddings.
- Responsibility area · 03
Documented focus: indexing.
- Responsibility area · 04
Documented focus: retrieval.
- Responsibility area · 05
Documented focus: reranking.
- Responsibility area · 06
Documented focus: RAG evaluation.
- Responsibility area · 07
Describe inputs, outputs, permissions and resources as a versioned capability.
- Responsibility area · 08
Keep relevant artifacts, provenance and status information traceable.
- Responsibility area · 09
Prepare handovers to adjacent Factories and cross-Lab capabilities through explicit contracts.
Factory-specific flow
How the RAG Factory fits the Orthevs model.
The flow connects the mission of the RAG Factory with the transitions defined by the Factory Contract. It is a target model for developing the capability, not evidence of an existing implementation.
- 01
Define the mission of the RAG Factory
The documented subject “Chunking, embeddings, indexing, retrieval, reranking, and RAG evaluation” establishes the domain starting point and boundary of the RAGF capability.
- 02
Describe inputs and outputs for RAGF
The manifest, schemas and versioned contracts are intended to state which inputs the RAG Factory processes and which outcomes or evidence it hands over.
- 03
Execute under the control of the AI & ML Engineering Lab
The Factory Manager, permissions, resource declarations and lifecycle states place execution under the responsible Lab Core.
- 04
Make RAG Factory outcomes traceable
Status, artifacts and provenance are intended to keep the flows and handovers relevant to “Chunking, embeddings, indexing, retrieval, reranking, and RAG evaluation” verifiable.
- 05
Verify contracts and boundaries
The RAGF capability remains within its documented ownership and uses adjacent Factories or platform services only through declared contracts.
- 06
Evolve through explicit versions
Changes to the mission, interfaces, artifacts or dependencies of the RAG Factory are intended to be maintained as traceable Factory versions.
Architecture role
How the RAG Factory is anchored in Orthevs.
The architecture separates the domain responsibility of the RAG Factory from shared Lab and platform services. This keeps the capability installable, versioned and integrable through contracts.
Domain context
The RAG Factory belongs to the “Generative AI & LLM Engineering” domain in the AI & ML Engineering Lab. Its scope is set by the Factory record and the focus areas named there.
Capability and lifecycle
The RAG Factory describes a specialist capability of the AMEL. Its Contract is intended to define inputs, outputs, permissions, versions and evidence. Depending on the need, it may be implemented as a plugin, package, service or adapter and qualified for a specific engineering process.
Interfaces and contracts
The RAGF capability is intended to declare identity, lifecycle, capability interfaces, artifacts, execution constraints and dependencies through versioned Factory Contracts.
Artifacts and evidence
The manifest, contracts, schemas, implementation and checks form the technical shape planned for the RAG Factory; concrete details become binding only with its Factory version.
Deliverables
What is planned to be versioned for the RAG Factory.
The artifact structure makes the mission of the RAG Factory tangible as a verifiable Factory capability. It describes the expected form of the work, not files or releases that already exist.
- 01
factory.manifest.yaml for identity, version, runtime, permissions and dependencies.
- 02
Versioned API, event and capability contracts.
- 03
Typed schemas for inputs, outputs, evidence and provenance.
- 04
Implementation, adapters and runtime integration.
- 05
Contract, lifecycle, security and integration tests.
Integration
Binding Factory Contract
The Factory Contract makes the capability installable, composable and controllable. It covers not only the interface, but also identity, lifecycle, execution constraints, security, observability, dependencies and compatibility.
- Identity and manifest with Factory ID, version, maintainer and trust origin.
- Lifecycle with installation, initialization, start, stop, upgrade and health states.
- Versioned capability interfaces including errors, timeouts and compatibility.
- Artifact contracts for schemas, versions, checksums and provenance.
- Execution constraints for runtime, compute, storage, scaling and timeouts.
- Security rules for permissions, secrets, isolation and audit events.
- Observability through logs, metrics, traces, health and correlation IDs.
The listed contract areas represent the minimum scope intended by the architecture concept. Concrete protocols, schemas and versioning rules are defined for each Factory implementation.
Quality, security and boundaries
Controlled responsibility for RAGF.
The RAG Factory has more than a domain mission. The Factory Contract also requires traceable permissions, execution boundaries, observability and documented handovers.
Boundary
The Factory stays within its declared domain boundary and uses platform-wide identity, secrets, policy, audit, events and artifact storage only through documented Orthevs contracts.
Security requirement
Security rules for permissions, secrets, isolation and audit events.
Observability and evidence
Observability through logs, metrics, traces, health and correlation IDs.
Maturity and roadmap
Planning state of the RAG Factory.
The RAG Factory is documented with the status “Concept in the Orthevs Factory catalog” and roadmap position “Phase 3 - Generative AI, priority P2”. This locates its target model in the portfolio; it does not establish a production scope or delivery date.
Traceable source basis
Derived directly from the Factory concept.
The mission, responsibility areas, boundaries, artifacts, contracts and roadmap on this page come from the versioned profile of the RAG Factory (RAGF) and the architecture concept of the AI & ML Engineering Lab.
Orthevs Factory Portfolio
Think about capabilities in context.
Factories connect through shared contracts to form traceable engineering flows — from domain design to operation.