LIGHTGBM GBDT ENGINE NDCG@10 = 0.9989 SUB-1MS EDGE INFERENCE $0.00015 / TASK

MinTok 1 Keywords: Edge-Hosted Predictive SEO

Extracted from the high-performance search intelligence pipeline in seo-alpha-site, MinTok-1-Keywords deploys a 45-tree LightGBM LambdaRank gradient-boosted decision tree directly into Cloudflare V8 worker isolates. It analyzes linguistic structure, commercial purchase intent, search volume potential, and SERP competition to predict true Expected Value (eV) before you write a single line of content.

NDCG@10 Ranking Accuracy

0.9989

LambdaRank v2 benchmark

Inference Latency

< 0.8ms

zero GPU cold start

Memory Footprint

1.4 MB

fits in 128MB isolate

Task Cost

$0.00015

100x cheaper than LLMs

TRY INTERACTIVE CALCULATOR GET API ACCESS TITLE OPTIMIZER →

Expected Value (eV) Mathematical Derivation

Traditional SEO focuses only on search volume, resulting in wasted engineering effort on low-conversion keywords. The MinTok-1-Keywords model calculates composite Expected Value (eV) by jointly optimizing commercial monetization potential against SERP ranking friction:

CALIBRATED FORMULA (LEAN LAMBDARANK V2)

eV = (Volume^0.5 × (CPC + 0.01)^0.5) / max(Competition^1.0, 0.01)

α = 0.5 (volume elasticity) // β = 0.5 (monetization weight) // γ = 1.0 (competition penalty)

Tier 1: Ultra High Opportunity eV > 350.0

Tier 2: High Value Target 180.0 ≤ eV ≤ 350.0

Tier 3: Standard Growth Vector eV < 180.0

13-Dimensional Structural Feature Vector

Before traversing the 45 gradient-boosted decision trees, queries are tokenized into 13 calibrated structural and semantic signals:

word_count & char_len: Head vs long-tail detection
avg_word_len: Linguistic and technical complexity
has_year & has_number: Recency and temporal specificity
is_question: Informational search intent flag
comm_density: Pricing, quote, comparison, ROI terms
trans_density: Signup, download, install, API terms
has_geo: Local, national, or regional search modifier
title_overlap: Structural title-to-keyword synergy
slug_overlap: URL path alignment index
content_words: Content depth and topical coverage

API Usage & Integration

Available via dedicated scoring endpoint and standard OpenAI SDK completions.

POST /v1/keywords/score DIRECT JSON

curl -X POST https://mintok.adstim.net/v1/keywords/score \
  -H "Authorization: Bearer mtk_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "keyword": "programmatic seo django cloudflare",
    "title": "Programmatic SEO with Django and Cloudflare",
    "slug": "programmatic-seo-django-cloudflare"
  }'

POST /v1/chat/completions OPENAI SDK

from openai import OpenAI

client = OpenAI(
    base_url="https://mintok.adstim.net/v1",
    api_key="mtk_live_..."
)

response = client.chat.completions.create(
    model="mintok-1-keywords",
    messages=[{"role": "user", "content": "programmatic seo"}]
)
print(response.choices[0].message.content)