PNS Research

ProblemPrivate-market diligence is limited by relevance, source discipline, and repeatable risk classification — not by access to text. The hard part is turning fragmented diligence materials into evidence-backed risk briefs that another analyst could reproduce.

SolutionSectorRisk Agent is a local, deterministic engine that ingests uploaded documents, retrieves evidence by domain-specific risk category, classifies evidence polarity, separates management claims from source-backed evidence, flags deterministic contradictions, then scores severity and confidence. It produces risk tables, generated diligence questions, and a cited Markdown memo with CSV / JSON / memo / audit-log exports that preserve the full evidence trail.

Implementation / progressArchitecture is Core Engine → Domain Packs → Workflow → Outputs. Retrieval is deterministic TF-IDF over PDF / TXT / CSV / JSON / MD inputs; upstream_energy is the complete first pack with a ten-category taxonomy, and chic_human_capital is a stub for future calibration. Python · Streamlit, no paid APIs. Where evidence is missing, the system reports “insufficient evidence” rather than inferring a finding, and every output requires human review.