Process Of Elimination Master

POEM

The First Reductive Inference Model

Most queries don't need an LLM.
They need the right answer, fast, honestly delivered.

POEM is a RIM — a Reductive Inference Model. Not AI.
A parallel paradigm built for the questions you ask every day.

88% Accuracy vs TinyLlama 74%

95.5× Faster 40ms vs 3,879ms

100× Smaller 10.9M vs 1.1B params

3 Compute steps vs full forward pass

Try POEM

Why POEM

You use an LLM to drive to the supermarket.
POEM is built for the supermarket run.

Most daily queries are factual, causal, or definitional. What is something. How does something work. What causes something. Why something happens.

For these — the vast majority of real daily usage — you don't need a billion-parameter model spending significant energy to generate a confident answer that may not be true. You need the right answer, fast, from a system that admits when it doesn't know.

Use an LLM when you need one. Use POEM when you don't.

What is photosynthesis? What causes earthquakes? How does gravity work? What is democracy? Why is the sky blue? What is machine learning? How does the immune system work? What causes climate change?

Two paradigms. Neither derived from the other.

Artificial Intelligence

Generates from parameters

Learns patterns from vast data, encodes them into billions of parameters, and generates answers by predicting what comes next. Powerful, general, expressive.

Best for: creative tasks, open-ended questions,
language understanding, complex reasoning

Reductive Inference

Eliminates to the answer

Classifies the question, eliminates what cannot be true, retrieves the answer from structured knowledge. If confidence is insufficient — it says so.

Best for: factual queries, daily knowledge,
speed-critical applications, honest retrieval

How POEM works

Classify the question

A trained neural classifier identifies the question type across 10 categories — factual, causal, definitional, comparative and more.

Eliminate the impossible

Nine wrong categories are eliminated immediately. The search runs only against what remains — fast, targeted, efficient.

Return or admit

The highest-confidence match is returned. If confidence is insufficient — POEM says it doesn't know rather than hallucinate.

Honest comparison

Different tools. Different strengths.
Not rivals.

Where POEM is better

Speed — 95.5× faster per query

Efficiency — 100× fewer parameters

Transparency — every answer is traceable

Honesty — admits uncertainty, never hallucinate

Cost — fraction of LLM inference cost

Updatable — knowledge grows without retraining

Where LLMs are better

Language understanding and generation

Open-ended and creative tasks

Context-dependent questions

Personal and location-aware queries

Reasoning across multiple knowledge pieces

Questions without a definitive answer

Benchmark — POEM vs TinyLlama 1.1B · 100 questions · 10 categories

88% accuracy. 95.5× faster. 3 compute steps.
For daily knowledge queries, you don't need an LLM.
RI and AI are parallel paradigms.

10.9M parameters · 246,000 knowledge entries · Built in Greece, 2026