Customer Evidence Packet Builder

A small applied AI workflow proof that turns messy customer evidence into a concise decision packet.

It is built for Forward Deployed Engineer, Applied AI Engineer, AI Product Engineer, and Solutions Engineer positioning: the work starts with ambiguous customer material and ends with a packet a team can actually use.

What It Does

• Ingests messy notes, source snippets, objections, goals, links, and implementation facts
• Supports JSON input or loose pasted notes
• Scores evidence readiness with deterministic, inspectable heuristics
• Builds audience-specific packets for client, internal, investor, or technical review
• Separates customer goals, top evidence, objections, implementation plan, open questions, and source map
• Exports markdown or JSON
• Runs with no external dependencies and no paid APIs

Quick Start

npm run dev

Open:

http://localhost:4174

Build a packet from the sample data:

npm run build

Build JSON for a technical review:

npm run build:json

Run the smoke check:

npm run check

Demo Flow

1. Click Load Demo.
2. Switch the audience from Client to Technical Review.
3. Review the recommendation, risks, implementation move, and source map.
4. Paste loose evidence with prefixes like Goal:, Objection:, Fact:, and Source:.
5. Copy the markdown packet.

Input Shape

The builder accepts structured JSON:

{
  "account": "Northstar Health",
  "project": "AI-assisted patient intake triage",
  "decision": "Should we run a scoped pilot?",
  "goals": ["Reduce manual intake review time"],
  "objections": ["Security lead is concerned about PHI retention"],
  "implementationFacts": ["Pilot can start with read-only CSV import"],
  "sourceSnippets": ["Screenshot shows 180 open queue items"],
  "evidenceItems": [
    {
      "title": "VP Ops discovery call",
      "source": "June 12 notes",
      "confidence": "high",
      "text": "Manual handoff blocks coordinators every morning."
    }
  ]
}

It also accepts loose pasted notes:

Goal: Reduce manual intake review time
Objection: Security lead is concerned about PHI retention
Fact: Pilot can start with read-only CSV import
Source: Screenshot shows 180 open queue items
VP Ops said the workflow is slow and blocks coordinators every morning.

Project Shape

index.html                    Browser app shell
src/evidence-engine.js        Deterministic packet builder
src/app.js                    Browser app wiring
src/styles.css                Product UI
data/sample-evidence.json     CLI demo data
data/sample-evidence.js       Browser demo data
scripts/build-packet.mjs      CLI packet builder
scripts/serve.mjs             Zero-dependency local server
scripts/check.mjs             Smoke check
docs/product-brief.md         Product and role-positioning brief
docs/app-screenshot.jpg       README screenshot

Why This Is FDE-Shaped

The repo demonstrates the customer-facing builder loop:

• collect messy evidence
• identify signal and risk
• translate stakeholder material into a clear decision
• keep source material visible
• produce the next implementation move
• leave a path for a model layer without requiring one for the proof

Next Upgrades

• Add local file import and export
• Add per-audience packet templates
• Add source attachment previews
• Add a lightweight review history
• Add optional LLM extraction behind the deterministic builder
• Add CRM, Notion, or Linear export adapters

Positioning Line

I build customer-facing AI tools that turn ambiguous evidence into decisions, pilots, and implementation plans.