How QCE fits

Between your data
and the AI that reasons over it.

📁
Your data
Millions of tokens
Multi-source · Multi-speaker
all records
Weavi QCE
Evidence selection
Deterministic · Zero-retention
Removes 50–99% of input
high-value evidence only
🧠
Your AI model
Focused reasoning
Any LLM · No lock-in
50–99%
Observed input reduction range
892 data points · 15 Sep 2026
50–53%
Paired model-token reduction
Sonnet 5 · Gemini 3.8 · Opus 5
87–97%
Paired cost reduction · real SEC
Medium & large corpus diagnostics
91.5%
Recall retained · physical 5M
With 80% input reduction

Range observed across 892 data points spanning 6 public semantic domains, real SEC EDGAR filings and physical scale tests from 500K to 5M tokens. Savings vary significantly by workload: compact or low-density corpora see smaller reductions; large, multi-source, governed archives see the highest. All figures are evidence observations from our test programme, not guarantees.

Your AI is reading everything.
It should only read what matters.

Organisations — government, healthcare, finance, legal, defence — hold vast data repositories. When an LLM is asked a question, it shouldn't wade through millions of irrelevant tokens. It should receive only the precise, relevant data it needs.

Without Weavi-QCE

Needle lost in the haystack

Sending an entire large corpus may exceed model context, increase input cost and expose the model to avoidable noise. Conventional pipelines therefore have to retrieve, chunk or truncate before reasoning.

With Weavi-QCE

Preserve the strongest signal

Your records go through the QCE first. It selects a budget-aware subset using declared importance, configuration and corpus signals, then returns full records for the downstream model. Realised reduction and quality are use-case dependent.

FDA adverse event data.
A live demonstration.

We built a reference application on top of the FDA's public adverse event database — millions of medical reports. A user queries: "aspirin bleeding elderly" and asks the LLM: "What are the risk factors?"

Without Weavi-QCE

Baseline pipeline

The reference application fetches an OpenFDA cohort and asks the selected LLM a user-supplied question. Large cohorts may require truncation or another retrieval layer before the model can process them.

With Weavi-QCE

QCE-assisted pipeline

The same model and question are run over QCE-selected, unmodified reports. The application displays both answers and token usage for inspection. This demonstrates integration and efficiency; it is not a clinical-validation claim.

01 — SUPPLY

Your authorised candidate records

Send documents, messages or structured records with token estimates and metadata. Current public requests are bounded; large-archive discovery and connectors remain a separate retrieval stage.

02 — FIND

Question-aware evidence selection

Configured variants use importance and corpus signals; advanced QCE variants also read the query, content, source, speaker and structural signals across the corpus. Quality remains task- and corpus-dependent.

03 — RETURN

Exact data, not summaries

We return the actual records — unmodified, unparaphrased. Your LLM gets the truth, not an interpretation of it.

04 — MEASURE

Quality and efficiency together

Measure retained tokens, recall, ranking quality, latency and downstream answer behaviour. Results depend on the corpus, configuration, budget and task.

Less context can mean
lower cost and better focus.

Matched 500K no-assistance calls: same query, same records, same 256K budget, no facets or retries. Models tested via OpenRouter gateway. Six-domain public datasets (science, health, finance, policy, climate, public-interest). Pair counts 6–8 per model.

Claude Sonnet 5
circa 50–53%
observed model-token reduction · recall gain of around 18 pts (8 pairs)
Gemini 3.8 Flash
circa 50–53%
observed model-token reduction · recall within 1.25 pts (8 pairs)
Claude Opus 5
circa 50–53%
observed model-token reduction · recall gain of around 17 pts (6 pairs)
Test data
Six public semantic domains + real SEC
Published labels span science, health, finance, policy, public-interest and climate data. Physical scale volume is disclosed separately.
Statistical rigour
Broad statistical and diagnostic suite
Paired tests, ranking metrics, effect estimates, resampling and multiple-comparison diagnostics are implemented. Reference-validation and independent-label improvements are on the scientific roadmap.
Retrieval metrics
Established retrieval metrics
MAP, MRR, NDCG@3/5/10, Precision@K, Recall@K and F1 provide complementary views of selection and ranking quality.
Reproducibility
21,130 recorded arm outcomes
Create-once manifests, frozen hashes, raw selected IDs, failures and statistical outputs are available for technical diligence. Request review →
See matched-call context and limitations
At physical 5M, full Sonnet 5, Gemini 3.8 Flash and Opus 5 arms were context-ineligible. QCE reduced the corpus to a 1M declared budget; Gemini's slightly larger context admitted 8/10 semantic calls, while Sonnet and Opus remained ineligible after overhead. Invalid calls were not retried or scored as misses. Latest model results are internal evidence for diligence, not regulatory approval or guaranteed customer outcomes.

From aircraft manuals to national security.
One API. Every scale.

Weavi-QCE is purpose-built for organisations with large, complex data where retrieval quality and model input cost can both be measured.

INSURANCE — CLAIMS & FRAUD

Multi-source claim evidence

Bring together policy documents, adjuster notes, calls, medical evidence and investigation records, then select a budgeted evidence pack for a specific coverage, fraud or complaint question.

AEROSPACE

Maintenance-manual retrieval

Prioritise relevant procedures across large technical-document corpora before sending the selected evidence to an engineer-facing model. Pilot outcomes should measure recall, lookup time and safety-critical omissions.

GOVERNMENT — TAX ADMINISTRATION

Investigation and case triage

Use configured materiality signals to reduce a large administrative corpus before semantic analysis and LLM review. Value depends on recall, analyst throughput and the cost of missed cases.

DEFENCE — INTELLIGENCE FUSION

Multi-source evidence selection

Combine declared importance and source metadata across heterogeneous records, selecting a reviewable subset for downstream analysis. Secure deployment and independent mission testing would be required.

LEGAL — E-DISCOVERY

E-discovery context selection

Reduce large email and document sets to a budgeted review corpus while preserving full records. A pilot should compare QCE with sparse, dense and hybrid retrieval under the same relevance labels.

HEALTHCARE — EHR

Longitudinal-record support

Select configured high-value records for authorised clinical review without summarising the source data. Healthcare deployment would require customer validation, privacy review and appropriate compliance controls.

AI PLATFORMS — WORKSPACE

Embedded context infrastructure

Data, cloud and model platforms could use a selection layer to manage long enterprise histories before inference. Distribution economics require partner-side volume and pricing validation.

SPACE — LAUNCH SAFETY

Safety-evidence preparation

Prepare a traceable subset of telemetry, inspections, historical events and weather evidence for expert and model review. Any safety claim requires domain-owner testing and governance.

LEGAL AI — CASE RESEARCH

Case-law evidence selection

Select full passages from large case-law corpora under a fixed model-input budget. Quality should be judged against independently labelled authorities and citation accuracy.

MORTGAGE FINANCE — UNDERWRITING

Multi-document loan analysis

A mid-tier bank processing 50,000 applications/year holds 100M+ tokens: applications, valuations, credit reports and compliance documents. QCE selects only what matters for each risk query — fraud signals, income evidence, covenant breaches — before any model sees it.

UK / EU / APAC GOVERNMENT

Regulated evidence at national scale

Tax authorities, central banks, regulatory bodies and public-sector AI programmes all hold massive, governed corpora. QCE enables LLM-powered policy review, compliance monitoring and citizen-service AI within strict data-minimisation constraints.

One selection layer.
Small context to physical multi-million-token evidence.

Weavi-QCE has been exercised from compact API requests through a 5M physical-token direct-selector study and 500M represented pressure. Physical, declared and represented quantities are reported separately.

Token scaleCurrent routeIllustrative workloadEvidence status
2 Ki – 32 Kilive APIChats, focused documents, agent memoryContract and internal testing
32 Ki – 128 Kilive APICases, claims, meetings, repositoriesInternal testing
128 Ki – 500 Kisix-domain dataRegulatory, medical, scientific and policy records96.5% latest labelled recall at 500K→256K
500 Ki – 5 Miphysical direct selectorLarge supplied candidate corpora91.5% latest labelled recall at physical 5M→1M
5 Mi – 500MrepresentedExtreme compression-ratio and budget-pressure studies62.5% recall at represented 500M→5M; not physical 500M

Test date: 15 Sep 2026. The physical-5M arm used 5,000 complete records and 20M content characters; neutral padding contributed parsing volume and is not independent semantic diversity. Represented scale exercises ranking and budget pressure on bounded physical text. Genuine massive archives require streaming/indexed shards and measured reduction equivalence.

"The best context window is not the biggest one — it is the most truthful one."

Weavi-QCE does not summarise or paraphrase. It selects. What reaches your model is drawn directly from your data — scored, ranked and preserved.

We believe confidence should come from inspection: testing routes, methodology, seeds and outputs are available for technical diligence. We welcome independent review and intend to publish a stronger reproduction package. Request access at hello@weavi.ai.

Ready to test the QCE on a real workflow?

Define the corpus, budget and success criteria first—then compare the result with your existing pipeline.

Customer portal View investor deck