Desktop support
Mobile support - chat
Guardrails towards experiences that does not exist

ExitWise

Don't let expertise retire with the expert.

ConHacks 2026 · exitwise.app · by team What?

What It Is

ExitWise is an AI-guided tacit knowledge extraction platform built for the six-month pre-retirement window. It conducts structured, multi-session interviews with retiring employees using an AI that probes — detecting vague answers, buried stories, skipped steps, and exception clauses — and converts those conversations into a structured, queryable knowledge profile for the successor.

Not a chatbot. Not a documentation tool. Not an exit interview. An intelligent extraction engine.

The Problem

80% of organizational knowledge is tacit. It walks out the door when people retire. Exit interviews are generic and ignored. Knowledge platforms can't extract what was never written down. Canada has 700,000 skilled trades workers retiring by 2028 and no adequate tool for the knowledge underneath them.

Three Users

Admin (HR/dept head) — sets up transfer engagements, pairs retiree + successor, sets domain/industry
Retiree — six structured sessions, 30–45 min each, over 8–12 weeks; receives personal legacy document
Successor — queryable knowledge profile; plain-language queries return the retiree's actual words

Session Arc

Trust + orientation — career in their own words
Critical processes — what breaks immediately without them
Decision-making under ambiguity — how they think
Relationships + org navigation — informal org chart
Edge cases + emergencies — rare, high-stakes scenarios
Review + refinement — retiree confirms the profile

AI Probe Signals

Engine watches for four patterns and probes before moving on:

Vague qualifiers — "you just know," "it depends" → what specifically
Skipped steps — process described too smoothly → what's between A and C
Buried stories — offhand incident reference → stop and ask
Exception clauses — "except when," "unless" → probe the exception

Stack

Frontend: React 19 · TypeScript · Vite · Tailwind CSS · Zustand · React Router v6 · React Hook Form + Zod

Backend: Hono · Bun runtime · Drizzle ORM · Custom JWT auth · GEMINI API (gemini-2.5-flash) · Server-Sent Events (streaming)

Infrastructure: Railway (hosting + API + PostgreSQL) · Vercel DNS (exitwise.app) · GitHub Actions CI/CD

Database

Six PostgreSQL tables: organizations, users, transfer engagements, interview sessions, exchanges, knowledge profiles.

running_summary JSONB column on sessions stores compressed conversation context — read once, passed to Gemini, written back after each exchange. AI maintains full conversation memory without reconstructing from scratch per request.

Key Architecture Notes

Streaming: React frontend, token-by-token render without blocking UI, session context updated in DB simultaneously
Session context injection: system prompt architecture + four signal triggers = most of the intellectual work
Trust architecture: retirees control release date, org cannot see content before release, retiree receives legacy doc in return

Design Language

Warm editorial aesthetic: deep forest green (green-deep), cream, amber. Communicates gravitas and dignity — critical for retiree trust. Square buttons (rounded-none), neo-brutalist shadows.