AI systems & infrastructure · v0.5.1 released

ARMIE Retrieval Platform

A production-oriented retrieval systems platform combining multiple retrieval strategies, deterministic constraint execution, observability, reproducible evaluation and a governed natural-language interface.

Probabilistic interpretation does not own deterministic retrieval authority.Language can propose intent; only confirmed, supported contracts can authorize execution.
Maturityv0.5.1 · Public GitHub release
RuntimeGoverned natural language → RetrievalContract → deterministic C1 → Elasticsearch
Validation10K controlled synthetic profiles · 120 benchmark queries · 16/16 live Founder-critical scenarios
BoundaryControlled synthetic evidence — not real-world expert-search validation

Retrieval becomes an AI systems problem when relevance, constraints and uncertain language meet execution.

A useful retrieval system must combine semantic relevance with explicit constraints while preventing uncertain language interpretation from silently removing valid candidates. ARMIE separates language interpretation, user authority and deterministic retrieval execution as distinct responsibilities.

Engineering principle. Retrieval is not a vector-search call. It is a system of contracts, runtime decisions, evidence and evaluation.

Retrieval as a system

Platform substrate

Query → Plan → Runtime → Result

Query declarative request
RetrievalPlan strategy, processors and constraints
RetrievalRuntime replaceable providers and registries
Dense · Sparse / Keyword · Hybrid / RRF · Graph retrieval strategies
Processors / reranking → RetrievalResult trace and evaluation remain visible
Product layer · v0.5.1

Governed interpretation → C1 authority

Natural-language request probabilistic input
Bounded interpretation → clarification → user resolution uncertainty remains explicit
Confirmation → RetrievalContract canonical execution boundary
Deterministic validation / C1 → Elasticsearch native constraints and provenance
Results + provenance same contract as manual structured C1

The v0.5.1 language layer sits above the retrieval platform; it does not replace the platform.

Natural language can propose intent. It cannot silently authorize execution.

Ambiguous supported intent triggers clarification. The user resolves the bounded choice, confirms the canonical contract, and only then can deterministic C1 execute. Unsupported meaning stays unsupported; editing a confirmed interpretation invalidates previous authorization. Manual structured C1 and confirmed natural-language execution converge at the same RetrievalContract boundary. No model participates in final compilation.

InterpretationClarificationUser resolutionConfirmationRetrievalContractC1

When automation failed held-out evaluation, the architecture changed.

Rule-based semantic-role classification looked promising on small development fixtures, then failed to generalize on a prospective held-out benchmark. A stronger local model improved parts of the task but did not make false hardening safe enough. The project did not promote automation simply because an LLM was available; uncertainty became an explicit product state routed back to the user.

Automation hypothesisDevelopment evidenceHeld-out failureStronger model testedFalse-HARD risk remainedClarification + user resolution
Use models where evidence supports them. Keep deterministic contracts around authority-sensitive execution.Evaluation was allowed to reject the architecture hypothesis.

Runtime and validation proof

10KControlled synthetic expert profiles
120Versioned benchmark queries
16 / 16Live Founder-critical Playwright scenarios
Elasticsearch 8.15.3Real constrained runtime
BGE-M3 · 1024-dDense embedding path
v0.5.1Public GitHub release

The benchmark is controlled synthetic evidence, not customer-scale or real-world expert-search quality. Current CI covers Python 3.9 and 3.11.

Constraint-aware execution

Native hard-constraint filtering

Supported C1 constraints use native Elasticsearch pre-filtering.

Strict shortfall

If fewer eligible results exist than requested, the system reports the shortfall rather than backfilling ineligible candidates.

Explicit user authority

Natural-language-derived hard constraints require clarification and confirmation where needed.

Provenance

Interpretation, contract and execution lineage remain inspectable.

Multiple strategies

Dense, Sparse/Keyword, Hybrid/RRF and Graph remain explicit platform strategies.

Observable processing

Reranking, result processing, traces and evaluation remain separate from planning and execution.

Designed beyond one retrieval domain

Expert Discovery is the released product scenario, but the core runtime uses domain-neutral objects and contracts such as Query, RetrievalPlan, RetrievalRuntime, ResultItem, RetrievalResult, registries and evaluation. A read-only architecture review found the generic platform suitable for a bounded second-domain proof without changing the core platform.

Boundary. Companion Memory integration has not shipped; this is architecture-readiness evidence only.

Current maturity and limitations

A retrieval platform was built, measured, stress-tested and deliberately constrained where evidence showed automation was unsafe. v0.5.1 includes the governed natural-language workflow, deterministic RetrievalContract/C1 execution, real Elasticsearch validation, Workbench integration, browser E2E and controlled evaluation infrastructure.

It does not claim real-world expert-network quality, unrestricted natural-language understanding, arbitrary automatic hard-constraint extraction, production multi-tenant deployment, universal cross-domain semantics or shipped Companion Memory integration.

Related direction

Could retrieval become a traceable foundation for long-term personal memory?

Explore Superbrain ↗