Retrieval & applications
Existing components support document retrieval and source inspection. ParancU iOS runs retrieval locally after corpus preparation, which sends source chunks to OpenAI using a user-supplied API key.
iOS privacy policy →Approach
Retrieve. Inspect.
Assess.
ParancU Lab is an independent applied AI research and development initiative investigating how to find evidence in documents and assess what it can support.
Semantic similarity can identify relevant passages. It does not guarantee that they contain the facts needed to answer a question.
An experimental approach
- 01 / RETRIEVE
Retrieve
Identify potentially relevant source passages using embedding-based retrieval and experimental retrieval strategies.
- 02 / INSPECT
Inspect
Examine the original source text, factual relationships, conditions, negations and supporting context.
- 03 / ASSESS
Assess
Investigate whether retrieved information is sufficient to answer a question, including missing evidence and potentially misleading passages.
These stages describe the laboratory’s research approach, not a fully validated end-to-end capability. Evidence-sufficiency evaluation remains experimental and is not an integrated feature of ParancU iOS.
Measure successes.
Study failures.
We evaluate retrieval and evidence-assessment approaches through controlled experiments, examining:
- Semantic relevance versus factual answer containment.
- Factual representations and question-oriented retrieval.
- Evidence completeness and missing information.
- Verification limitations and unsupported answers.
- Source traceability and preservation of original meaning.
Experimental findings document what was observed under specific test conditions, not general guarantees of reliability.
Research maturity & processing boundaries
Agent Lab workflows
Agent Lab explores evidence inspection, answer generation, verification and limited recovery. Verification does not guarantee correctness.
Imported prepared corpora support browser-local retrieval only. Separate server-assisted workflows send content to the server and use OpenAI for preparation, answers and verification.
Agent Lab data handling →Under investigation: ParancU 2.0 and ParancU Decision Lab explore new factual representations, question-oriented retrieval strategies, evidence sufficiency and complementary evidence selection.
Cloud/API integration, enterprise privacy architecture and commercial deployment models remain possible future directions. Document confidentiality is a design requirement, not an established guarantee.