Skip to content
Jackson Macdonald
← All work

AI Product · Decision Intelligence · Simulation

War Room AI

A dynasty draft copilot designed to answer: What is the best decision?

— not simply: “Who is the best player?”

War Room AI combines current league state, player data, market information, roster strategy, probabilistic simulation, and a deterministic Expected Draft Value framework before using AI to explain and challenge the recommendation.

simulations
10,000
Expected Draft Value framework
EDV
league integration
Sleeper

AI explanation layered on deterministic logic.

The product problem

Fantasy draft tools are usually optimized around player rankings. That is useful, but it does not fully answer the question a manager faces during a live draft.

At a specific pick, the real decision depends on who is available now, who may still be available later, roster needs, positional scarcity, tier cliffs, future draft capital, trade value, league format, and the opportunity cost of passing.

The product was designed around that decision.

Product philosophy

War Room AI does not ask: “Who is the best player?”

It asks: “What is the best decision at this specific pick?”

Every recommendation should explain

  • Why this player?
  • Why now?
  • Why not later?
  • What happens if I pass?
  • What should the next sequence become?

Core architecture

The four core intelligence layers

Scout Engine

Owns player intelligence.

  • Scouting summaries
  • Strengths / weaknesses
  • Ceiling / median / failure cases
  • Landing-spot analysis
  • Historical comparisons
  • Risk, confidence, sources

Facts and calculated comparisons should exist before AI explanation.

Simulator Engine

Runs 10,000 draft simulations.

  • Inputs: ADP, consensus rankings, NFL draft capital, team needs, positional runs, league settings, randomness
  • Outputs: survival probability, tier survival, board states, value pockets, contingency scenarios

Key question: if I pass now, how likely am I to get this player or a comparable option later?

Strategist Engine

Knows the user's team.

  • Evaluates roster strengths, weaknesses, competitive window, positional scarcity, future draft capital
  • Produces pick priorities, draft philosophy, need vs. value balance, trade opportunities

Decision Engine

Calculates Expected Draft Value (EDV).

  • Factors: player value, roster fit, scarcity, tier urgency, upside, risk, reach penalty, survival probability, future value, alternative-player survival, trade value, future pick sequencing
  • Outputs: recommended decision, confidence, alternatives, decision tree, explanation factors

The most important architecture decision

The LLM does not decide the pick from scratch.

The quantitative / deterministic layer produces the underlying recommendation. The AI layer is used to explain, contextualize, compare, challenge, and answer natural-language questions.

This keeps current facts, probabilities, and decision logic separate from generative explanation.

AI explains the evidence. It does not invent the evidence.

Data model / grounding

Sleeper integration imports

  • League metadata
  • League settings
  • Rosters
  • Users
  • Draft order
  • Traded picks
  • Player identities

Player intelligence

  • Rookie ADP
  • Consensus rankings
  • Position rank
  • NFL draft capital
  • College production
  • Athletic testing
  • Age
  • Size
  • Landing spot
  • Depth chart
  • Injury status
  • Scouting research
  • Current news

Every current external data point retains provenance and freshness information.

The system should not rely on unverified AI memory for current prospect facts.

Live draft workflow

Every pick can change availability probabilities, positional scarcity, roster needs, future board states, recommendation, confidence, and trade signals.

Product experience

Draft prep

Primary screens

  • Tier Board
  • Pick Planner
  • Player Explorer
  • Simulation Center
  • AI Chat

Questions it should help answer

  • What happens if I wait?
  • Which tiers disappear before my next pick?
  • Where are the value pockets?
  • Can I trade back?
  • Which players fit my roster?
  • What are my contingency plans?

Draft day

Primary interface

  • Draft Board
  • Available Players
  • AI Recommendation
  • Future Planner

Supporting functionality

  • Manual pick entry
  • Undo
  • Autosave
  • State recovery
  • Recommendation refresh
  • Queue
  • Future availability
  • Scout cards
  • Trade-up / trade-back signals
  • Branch comparison

Technical architecture

Planned / scoped stack

  • Next.js
  • TypeScript
  • Tailwind CSS
  • shadcn/ui
  • Supabase
  • PostgreSQL
  • Prisma
  • OpenAI Responses API / AI integration layer
  • TypeScript and/or Python simulation logic

Designing beyond the happy path

Example end-to-end flow

  1. 1.Import league
  2. 2.Review roster
  3. 3.Rank player
  4. 4.Run simulation
  5. 5.Open pick planner
  6. 6.Start live draft
  7. 7.Record picks
  8. 8.Undo a pick
  9. 9.Receive recommendation
  10. 10.Refresh page and restore state

Golden scenario

My role

I conceived the product, defined the product philosophy, scoped the MVP, designed the core system architecture, worked through the decision framework, specified inputs and outputs, defined the major data structures, and used AI-assisted development workflows to move from concept toward implementation.

What this demonstrates

  • AI product design
  • Product strategy
  • Decision intelligence
  • Simulation
  • Data architecture
  • Requirements
  • Deterministic + generative AI
  • AI grounding
  • Live-state systems
  • Product UX
  • AI-assisted development
  • Edge-case thinking

Status

A personal AI product / MVP architecture and build project.

The 10,000 figure refers to simulation iterations within the scoped simulator, not users.

The interesting part wasn't adding a chatbot to fantasy football. It was separating the decision system from the AI explanation layer.

That architecture makes the recommendation inspectable, testable, and grounded in current data.

Building AI systems that have to work in the real world?

I’m interested in AI implementation, deployment, transformation, forward-deployed, and product-adjacent roles where I can own the path from business problem to deployed outcome.