Supplie - AI Decision Support Exploration

Background / Problem

Operational teams often rely on dashboards and spreadsheets to track what has already happened. These tools are useful, but they rarely help teams explore what might happen next.

The idea behind Supplie was to explore whether AI models could help surface useful signals from messy operational data and support planning decisions rather than just reporting.

The focus is not automation. The focus is helping humans understand complex systems faster.

Approach

I started by looking at how operational teams actually make decisions. A lot of the work happens in spreadsheets, Slack threads and meetings where people are trying to interpret partial information.

From there I explored a few directions:

• identifying signals that matter in supply chains and logistics
• experimenting with ways AI models could surface patterns or anomalies
• prototyping interfaces where teams could explore scenarios rather than just viewing reports
• exploring simulation-style approaches to planning

Most of the work so far has been early exploration and prototyping rather than a production system.

Working prototype

This came out of a couple of emails discussing a planning problem.

Rather than trying to explain the idea in words, I built a simple working prototype within a few hours to sanity check how it might behave.

Sharing something real made it much easier to align on what was useful, what didn’t make sense, and where the idea could go next.

Emails → quick prototype → feedback → iteration

Outcome / What I learned

A few interesting themes came out of the exploration.

• Operational data is messy. Real-world systems rarely have clean inputs.
• AI is most useful when it helps people reason about decisions rather than when it tries to replace the decision entirely.
• The most interesting opportunity may be helping teams explore scenarios rather than generating answers.

My role

• Concept exploration
• Product design
• AI experimentation and prototyping
• Research into operational decision workflows

Prototype-first development

Supplie is less about building a finished platform and more about exploring how AI could assist complex operational decisions.

Rather than starting with a large system design, the approach has been to build small working prototypes to test ideas quickly.

Instead of producing slide decks or theoretical models, the goal is to move quickly from discussion to something tangible.

A typical workflow looks like this:

• Identify an operational decision problem
• Sketch a potential AI-assisted workflow
• Build a small prototype to explore the idea
• Test the concept using real or simulated scenarios
• Refine the idea based on what proves useful

Even rough prototypes tend to generate far more useful discussion than conceptual presentations.

Tools used during exploration

Because Supplie is still in an experimentation phase, most work so far has focused on rapid prototyping rather than building a full production system.

Tools used during this exploration include:

• Claude and other LLM tools for reasoning experiments and concept validation
• Codex and AI-assisted development environments for rapid prototyping
• Vercel for quickly deploying experimental interfaces and prototype tools

These tools allow ideas to move from concept to working experiment quickly.

Example workflow being explored

Many of the ideas explored in Supplie follow a simple pattern:

Signals → reasoning → decision support.

Operational signals are collected and interpreted by an AI reasoning layer which attempts to surface meaningful changes in the system.

Examples of signals might include:

• inventory levels
• demand patterns
• lead times
• supply disruptions
• external constraints

Rather than simply displaying these signals in dashboards, the goal is to explore how AI could interpret them together and highlight where attention or intervention might be needed.

Why prototypes instead of slide decks

Operational systems are complex, and many ideas sound promising in theory but fail when tested.

Working prototypes make it easier to see:

• how signals interact
• what the AI actually produces
• whether the output is genuinely useful

This approach allows concepts to be validated before investing in larger systems.

Future system direction

If the most promising concepts from these experiments were developed further, the system would likely evolve toward a more structured architecture including:

• signal ingestion pipelines
• data models representing operational networks
• AI reasoning layers capable of analysing signal interactions
• simulation or forecasting components
• interfaces for operational teams to explore recommendations

The long-term goal is to move beyond passive reporting systems toward AI-assisted operational decision support.