Built by an engineering studio that has shipped since 2015.
Rank Tuner is made by EltexSoft — a software studio with 30+ senior developers on retainer. We built it to run our own content, then realized the discipline it enforces is the product: mine, research, write, gate, publish, measure, decide. Eleven years of shipping production systems taught us the hard part is never generation. It is judgment, and the loop that compounds it.
Same faces, month 1 and month 36
EltexSoft is a retained team, not a staffing churn. We have shipped for Google, Boeing, Shell, Teamworks, and Unfold across Laravel, React, Kotlin, Django, and Postgres. Rank Tuner runs on the same stack we deploy for clients, because we wanted one ops model we already trust.
Claude is the reasoning engine of the loop
A single frontier model does deep research with the web search tool, structured drafting into Portable Text, claim validation, and the daily analysis that steers the backlog. Prompt caching amortizes the static company and voice context across every article; structured outputs guarantee machine-parseable artifacts at every stage; and a provenance row captures model, prompt version, sources, tokens, and cost for every task. Claude is the product’s AI engine — the thing that does the reasoning the loop runs on.
What we will not compromise
The human holds the gate
Automation does the work; a person decides what ships. We will not build an auto-publish toggle, because the gate is the product’s integrity.
Statistical, not magic numbers
Decay, cannibalization, and link confidence are relative and tunable. No hardcoded thresholds pretending to be insight.
Every task is accountable
Model, prompt version, sources, tokens, and cost are logged for every run. Nothing fails silently; everything is auditable.