AI Workflow Mapper

Conversation-to-graph tool for mapping workflows, finding friction, and planning practical AI interventions.

Workflow impact: Designed to make AI adoption more concrete by turning vague process discussions into visible maps, ranked opportunities, and first experiments with human review points.

Project Description

I built this prototype to explore the core problem behind AI adoption: before you can automate a process, you need to understand how the work actually happens. The tool turns a casual conversation about a workflow into structured JSON, renders it as a node graph, then highlights friction points, handoffs, risk areas, and low-risk AI opportunities.

Example Enterprise Scenario

One built-in demo maps a client-data intake workflow: client records arrive through a shared email inbox, coordinators sort and classify them manually, analysts extract fields from attachments, exceptions loop back to the client, sensitive cases go through compliance review, and approved data is entered into CRM and operations systems. The map highlights where AI can safely help with classification, extraction, validation checklists, clarification drafts, and handoff packets while preserving human approval for risk-bearing decisions.

Key Features

Conversational workflow intake

Turns casual process descriptions into structured workflow steps, owners, handoffs, and open questions.

LLM-to-JSON process extraction

Uses runtime schema validation and can use structured outputs when an OpenAI API key is configured.

Node-based workflow graph

Renders tasks, decisions, handoffs, reviews, bottlenecks, inputs, outputs, and AI-assist points as a navigable map.

Friction/time/risk overlays

Switches graph emphasis across friction, time sinks, AI suitability, risk, and handoff complexity.

Step-through playback

Walks stakeholders through the process one node at a time with bottleneck and review cues.

Ranked AI intervention plan

Groups recommendations into quick wins, high-value projects, human-only areas, and discovery gaps.