Aletheia
Snapshot demo data layer
Source: /demo-data exports
Public demo mode

This hosted demo uses real precomputed outputs generated by the full Aletheia local pipeline.

Live retrieval, index rebuilds, and admin jobs are disabled publicly to avoid always-on search and ML infrastructure costs.

The full live stack runs locally with FastAPI, Postgres, Redis/RQ, OpenSearch, Qdrant, embeddings, and reranking.
Experiments

Experiment comparison matrix

Compare exported retrieval configs against real qrels-backed evaluation runs from the full local pipeline.

Snapshot source /demo-data/experiments.json

Public experiment snapshot

Best-by-metric indicators use real exported evaluation metrics. Admin comparison jobs are disabled publicly.

0 configs and 0 latest completed runs loaded.

Best Recall@10

Computed from latest completed runs.

Not enough comparable data.

Best MRR@10

Computed from latest completed runs.

Not enough comparable data.

Best NDCG@10

Computed from latest completed runs.

Not enough comparable data.

Fastest avg latency

Computed from latest completed runs.

Not enough comparable data.

Quality comparison

Latest completed runs with real Recall@10, MRR@10, and NDCG@10 values.

No real quality metrics are available for charting.

Latency comparison

Latest completed runs with real measured latency fields.

No real latency values are available for charting.

Comparison matrix

Latest completed evaluation run per experiment config. Missing values stay unavailable.

No experiment configs are available for comparison.

Experiment configs

Reusable retrieval parameter sets returned by FastAPI.

No experiment configs returned by FastAPI.

Select an experiment config

No config is selected.

Config details are shown only for real configs returned by FastAPI.

Public snapshot actions

Admin comparison jobs are intentionally disabled in public snapshot mode.

The experiment rows and metrics are real exported outputs. Run the full local stack to seed configs or launch new comparison jobs.