Chapter 9: Explainable Comparison

EggHatch-AI now includes a structured comparison layer for gaming laptop questions. This is one of the most meaningful recent changes in the repository because it moves the system from “giving recommendations” to explaining why one option wins over another.

Before this feature, the project could summarize trends, sentiment, and candidate laptops, but the final answer still depended heavily on free-form LLM wording. That made comparisons harder to trust.

The new comparison flow changes that by producing a deterministic comparison object before the final response is written.

Why This Matters

Laptop shopping is full of tradeoffs:

• price vs performance
• portability vs thermal headroom
• review quality vs raw specs
• value vs premium features

If the system only answers with prose, the user gets a recommendation but not necessarily a decision framework. The new comparison module gives EggHatch-AI a more explainable shape.

What The Comparison Layer Produces

The comparison helper generates a structured object that includes:

• the recommended laptop
• a short summary of why it won
• top reasons
• main tradeoffs
• ranked candidate laptops
• dimension winners

Those dimensions are intentionally simple and local to the current POC:

1. Performance
2. Value
3. Reviews
4. Portability

The Flow

flowchart LR
    Query["Comparison query"] --> Detect["Detect comparison intent"]
    Detect --> Candidates["Pick top 2-3 laptop candidates"]
    Candidates --> Score["Score performance, value, reviews, portability"]
    Score --> Rank["Rank candidates"]
    Rank --> Explain["Generate reasons and tradeoffs"]
    Explain --> UI["Render comparison breakdown in dashboard"]

Where It Lives In The Code

The deterministic logic is isolated in:

app/agents/comparison.py

That file is intentionally not a giant recommendation engine. It is a focused helper that keeps the explanation logic:

• small
• inspectable
• testable
• independent from the LLM wording layer

Why This Is A Good Prototype Design

EggHatch-AI is still a local prototype. That means it should not pretend to have production-grade ranking logic. The comparison helper reflects that philosophy:

• it uses fixture data
• it uses simple weighting instead of hidden models
• it degrades cleanly
• it makes tradeoffs visible

This makes it a better teaching artifact and a better open-source demo.