Mario Alexandre
I spent 7 years designing electrical systems where a mistake means a building goes dark. Then I applied that same engineering discipline to AI. The sinc-LLM framework is what happens when signal processing theory meets production software.
Why an Electrical Engineer Built a Prompt Framework
The connection is not accidental. It is the same discipline applied to a different medium.
In electrical engineering, the Nyquist-Shannon sampling theorem determines how many samples you need to faithfully reconstruct a signal. Sample too few times and you get aliasing: distorted, unreliable output that looks real but is fundamentally wrong. Every EE student learns this in their second year.
When I started building production AI systems and watching LLMs produce hedged, vague, structurally incoherent outputs, I recognized the pattern immediately. The model was not broken. The input was undersampled. A raw prompt is 1 sample of a 6-band specification signal. The model reconstructs the missing 5 bands from its training distribution, and that reconstruction produces aliasing artifacts: hedging, hallucination, generic responses.
The fix is the same fix that has worked in signal processing since 1949: sample at the Nyquist rate. For LLM prompts, that means providing all 6 specification bands. The math transfers directly. The zone functions I fitted to the empirical data use the same sigmoid, Gaussian, and ramp structures I studied in my BSEE program at the University of South Florida.
The sinc-LLM framework exists because I happen to have a foot in both worlds: the signal processing theory from my electrical engineering education, and the production software experience from building AI systems that handle real money and real clients. Most prompt engineers do not know what a sinc function is. Most signal processing engineers do not build LLM applications. I do both.
Career Timeline
What I Build
sinc-LLM
The Nyquist-Shannon sampling theorem applied to LLM prompts. 6-band specification decomposition, SNR computation, auto-scatter engine. 275 production observations, 97% cost reduction measured. Open source, MIT license.
AI Assistant System
Production AI assistant with layered cognitive architecture, contextual memory across sessions, multi-channel deployment (Telegram, iMessage), and autonomous action capabilities. Built with a post-response gate that detects when the AI asks questions it already knows the answers to and rewrites the response to act instead.
Content Automation Pipeline
End-to-end system that transforms raw video transcripts into publication-ready content: blog articles, social media carousels, short-form video with synchronized captions, B-roll, and hook overlays. Drives organic traffic and revenue for clients at scale.
Payment & Access Automation
Engineered payment systems connecting e-commerce platforms to CRM, handling subscriptions, access grants, and revocations with zero manual intervention. Processes real transactions daily.
Multi-Agent Orchestration System
Built an 11-agent system where each agent has a specialized role: code execution, creative exploration, quality evaluation, prompt optimization, edge detection, research, and more. All agents communicate through a shared autopoietic knowledge vault and follow the sinc-LLM specification format.
Browser Automation Infrastructure
Built browser automation infrastructure that handles platform operations previously requiring hours of manual data entry. Runs unattended with session management, authentication persistence, and error recovery.
Technical Skills
Languages
English. Native / Bilingual
Portuguese. Native / Bilingual
Spanish. Full Professional
Work With Me
I build custom sinc-formatted prompts for businesses that need their AI to perform. I also consult on AI automation infrastructure.
View Services