evidence-provenance

Local-first scientific literature RAG with quote-anchored evidence annotations, human paper screening, frozen provenance, and model-free research worksheets.

Evidence-governed outlet insight component for consumer-goods decision workflows; human-reviewed, no execution.

Scoped evidence benchmark: native Boolean pilot, synthetic replay, separate C/C4 provenance and reproducibility

Evidence-first product strategy research on consumer electronics discovery across four research surfaces.

Harness-neutral, evidence-first company research with versioned contracts, MCP adapters, and completed-only publication.

Where AI agents disagree before humans decide. A multi-agent risk investigation engine with evidence provenance, an adversarial red team and transparent scoring.

Research-only rare-disease evidence infrastructure, currently implemented and evaluated for TSC1/TSC2. Builds reproducible variant evidence packets, leakage-safe validation controls, and a source-grounded Mechanism Atlas. Not a clinical or diagnostic system; issues no authoritative variant classifications.

Active evidence assurance for warehouse robots: verify, repair, and gate perception before motion on AMD Radeon and ROCm.

Evidence-driven document intelligence for heterogeneous healthcare claim documents, turning PDF bundles into structured, traceable clinical information.

Local-first report and evidence curation with synthetic examples and source integrity checks.

Copy-aware evidence corroboration for offline retrieval-augmented verification

Offline synthetic security investigation with source-linked Security RAG retrieval and a narrated walkthrough

Evidence-bound Kubernetes identity incident investigation and tested remediation prototype

Local literature-review workspace: search, screening, citations, and claim-level evidence records.

EviGraph - AI-Powered Criminal Network Analysis & Cryptographic Evidence Provenance System (Smart India Hackathon 2026)

Local-first visual preference language prototype with pairwise comparison, evidence-linked axes, deterministic holdout feedback, and explicit uncertainty.

Deterministic simulation bench for runtime governance of Physical AI (AIBL), Phases 0–4.

Research prototype for AI-agent forensic evidence capture, dependency tracking, and trust calibration.

RISK//REPLAY -- AI Decision Forensics & Counterfactual Replay Engine. Deterministic replay, mutation, counterfactual, causal analysis, risk & governance scoring.