RAG optimization: reduced response time and LLM cost by 80%+
Average response time fell from 9.1 s to 1.6 s, and LLM cost from $4.03 to $0.46 per 1,000 questions. The difference in answer accuracy was too small to call either way.
A combination of 20+ years of technical experience and business know-how acquired through owning and running ventures fueled by Google Ads and Analytics.
Marty Paukstys, the founder of D2CEBL, has been operating his own businesses for over 20 years. He studied computer science, and for the last several years has been building and operating software products. The work spans product development, data, AI, automation and marketing, usually all at the same time.
Marty Paukstys on LinkedInAverage response time fell from 9.1 s to 1.6 s, and LLM cost from $4.03 to $0.46 per 1,000 questions. The difference in answer accuracy was too small to call either way.
$1,281,235 ad spend · 64,009 leads · $19.97 cost per lead · 40.04% conversion rate · $9,167,785 in sales · 716% return on ad spend
GPT‑6 Astra finished first in all 26 grading batches, averaging 4.1 minutes per 30 answers vs. GPT‑5.6 Sol’s 7.3. They agreed on 95% of verdicts, but Sol graded the answers more strictly.
Four live projects, with several more under development.
Measured in our RAG experiment on a 736‑page Ford F‑150 manual
Owning a business and managing its paid search on top provides a different perspective than running paid search campaigns for your clients. Cost-effectiveness and bottom-line results become much more relatable.
$1,281,235 Google Ads spend · 40.04% conversion rate · $19.97 cost per lead · 13.98% advertising cost
Read the studyAI is a powerful tool, but not everyone needs it. Not all businesses are the same, and AI is not always the right solution to every problem.
“I will not sell. I don’t want your business unless you think I can help yours.”
30 minute conversation to see if we are the right fit