Exploration
Discover a schema before committing to one. Use agent-driven clustering, sampling, pattern discovery, frame and role mining, and data hygiene checks.
SUPPORTING AI LAYER / HUMAN-IN-THE-LOOP MODEL DEVELOPMENT
Explore. Validate. Deploy.
Continuously harden.
Bring raw intelligence, expert annotation, model evaluation, and deployment into one reproducible workflow, with operational failures fed back into retraining.
THE PROBLEM IT SOLVESFragmented labeling tools, spreadsheets, training pipelines, and evaluation scripts make it difficult to turn operational model failures into lasting improvements.
Explore → Annotate → Evaluate → Improve
Annotation
Focus labeling on high-impact data with active-learning sampling, inline model suggestions, knowledge-linked entities, and keyboard-driven workflows.
Evaluation
Compare model outputs, correct them in context, track changes, and flag critical issues or preferences. Use inter-annotator agreement and learning curves to assess quality.
Continuous improvement
Detect systematic production errors, draft corrections, route them for expert validation, and queue retraining so models keep learning from the mission.
MINSKY CANDOR / IN PRACTICE
One agent. Two loops.
Candor is both an MCP server and client. The exploration loop helps discover entities and schemas; the improvement loop analyzes production errors and proposes fixes. People, services, and agents use the same operations through UI, API, and MCP.
Any source. Any task. Any backend.
Connect files, SQL, Elasticsearch, Neo4j, S3, or custom stores. Work across document, span, and frame tasks including classification, clustering, entities, relations, and semantic roles. Compare experiments across ML backends and LLM APIs.
Governed, auditable, searchable
Search and export predictions by label and confidence with data and schema provenance. Support role-based access, audit trails, project sign-off, and the needs of ML engineers, annotators, project leads, data consumers, and platform administrators.
SUPPORTING AI LAYER / BUILD & IMPROVE
Expert feedback becomes institutional learning.
Candor is a supporting AI layer that helps build, evaluate, and continuously improve the models behind our products. Minsky Platform supplies the shared knowledge foundation; IPE and STIO apply it to mission workflows. Candor is part of Minsky One.
YOUR DATA. YOUR MODELS. YOUR MISSION.
Bring your next
hard question.
Bring your use case. We’ll show you where dMetrics fits.