One-click agent runs that turn a dense deal data-room into a checklist report where every answer links to its source quote — plus memo drafting, rate tables, and data cleaning on the same architecture. Being sold to large reinsurers and asset managers to run deal intake — agentic workflows overseeing and accelerating every step of assessment.
Reinsurance is insurance for insurance companies. When an insurer shops risk to reinsurers, someone must work a roughly 59-field intake checklist through dozens of pages of dense documents before anyone can quote — hours of skilled reading and cross-referencing per deal.
The app converts the deal documents to text and launches a headless coding agent in a sandboxed workspace. The lead agent fans the checklist out to parallel subagents, each owning a group of fields; their structured results merge by script into a client-presentable report — every field with a found/partial/missing status badge, a source citation, and a verbatim supporting quote, plus severity-ranked red flags and ready-to-send open questions. A chat panel resumes the same agent session for follow-ups. Three more workflows — pricing memos, rate tables, data cleaning — reuse the architecture: AI does the judgment, deterministic code does the math.
A headless agent CLI as the engine with pluggable alternatives behind a model picker; parallel fan-out with cheap fast models drafting and a stronger model assembling; domain procedures packaged as reusable skills; grounding rules throughout — verbatim quotes as evidence and "silent in materials" instead of guessing.
Python, FastAPI, Claude Code CLI (headless, subagents, custom skills), NDJSON streaming, python-docx / pdfplumber / openpyxl, vanilla JS.