Track: ArcadeDB for Peregrine knowledge graph and Robin system graph (future) #54

New issue

Open

opened 2026-07-05 09:07:36 -07:00 by pyr0ball · 0 comments

pyr0ball commented

2026-07-05 09:07:36 -07:00

Owner

Source

https://github.com/ArcadeData/arcadedb — Apache 2.0

Context

ArcadeDB is a multi-model database (graph + document + vector + time-series + search)
with built-in MCP server. Active evals in Turnstone and cf-orch (see those tickets).
This ticket tracks two longer-horizon CF use cases.

Peregrine: job knowledge graph

Peregrine's job matching is currently document-centric (resume vs job description).
ArcadeDB graph model would enable:

Skill relationship graph: "Python" → relates to → "FastAPI", "data science",
"ML engineering" — traverse to find adjacent skills to recommend learning
Job market graph: company → industry → role family → skills required
User career graph: jobs held → skills demonstrated → skills inferred → gaps to target
Vector embeddings for semantic job/resume matching alongside graph traversal

Combined query: "find jobs semantically similar to user's target role AND where the
required skills are within 2 graph hops of skills the user already has."

Robin: Linux system knowledge graph

Robin (Linux migration companion) needs to know:

Windows app → Linux equivalent mappings (with quality/maturity ratings)
Package dependency graphs (install X, also needs Y and Z)
User's current system state as triples
Conflict relationships: app X conflicts with library Y version Z

ArcadeDB's graph + document model in one database is cleaner than a separate Oxigraph
instance (RDF) + SQLite (system state). The Cypher query interface is also more
approachable than SPARQL for this use case.

Priority

Now: Turnstone and cf-orch evals (see those tickets)
Next: Peregrine knowledge graph — activate when Peregrine enters v1.0 scope
Later: Robin — activate when Robin exits concept stage

Turnstone eval ticket
circuitforge-orch#87 (ArcadeDB for agent registry)
kiwi#155 (Oxigraph for food ontology — different tool, different scale)

## Source https://github.com/ArcadeData/arcadedb — Apache 2.0 ## Context ArcadeDB is a multi-model database (graph + document + vector + time-series + search) with built-in MCP server. Active evals in Turnstone and cf-orch (see those tickets). This ticket tracks two longer-horizon CF use cases. ## Peregrine: job knowledge graph Peregrine's job matching is currently document-centric (resume vs job description). ArcadeDB graph model would enable: - Skill relationship graph: "Python" → relates to → "FastAPI", "data science", "ML engineering" — traverse to find adjacent skills to recommend learning - Job market graph: company → industry → role family → skills required - User career graph: jobs held → skills demonstrated → skills inferred → gaps to target - Vector embeddings for semantic job/resume matching alongside graph traversal Combined query: "find jobs semantically similar to user's target role AND where the required skills are within 2 graph hops of skills the user already has." ## Robin: Linux system knowledge graph Robin (Linux migration companion) needs to know: - Windows app → Linux equivalent mappings (with quality/maturity ratings) - Package dependency graphs (install X, also needs Y and Z) - User's current system state as triples - Conflict relationships: app X conflicts with library Y version Z ArcadeDB's graph + document model in one database is cleaner than a separate Oxigraph instance (RDF) + SQLite (system state). The Cypher query interface is also more approachable than SPARQL for this use case. ## Priority - **Now:** Turnstone and cf-orch evals (see those tickets) - **Next:** Peregrine knowledge graph — activate when Peregrine enters v1.0 scope - **Later:** Robin — activate when Robin exits concept stage ## Related - Turnstone eval ticket - circuitforge-orch#87 (ArcadeDB for agent registry) - kiwi#155 (Oxigraph for food ontology — different tool, different scale)