? Oops! Sorry
The mobile version is not created.
I'm working.
Please visit on desktop version.
Project
Bardie
Date
Tasks
March 2026
Product DesignIT & Sales StrategyPrompt EngineeringRAG ArchitecturePHP BackendAI Fine-TuningVibe Coding
Link
Doc
Map
An Engineering Kerning LABS® Case Study
Engineering a production-ready AI sales agent, from zero to deployment in 7 days. I built Bardie, a proprietary RAG system and multi-tenant PHP infrastructure to prove that generative AI can autonomously execute complex sales methodologies.
#1 The Vision
Most AI systems are built to answer questions. Bardie was built to close deals. The vision behind this project was to close the gap between technical automation and human sales psychology — transforming a generative AI from a passive support tool into a high-performance commercial agent capable of executing complex methodologies like ZMOT and Three Gaps across multiple Buyer Personas simultaneously.
The result is a fully deployed Generative AI system (RAG) purpose-built for pre-booking automation, taken from concept to functional product in a 7-day development cycle. Built on a validated sales core powered by Groq, an asynchronous messaging architecture via SMTP and Cron Jobs, and a scalable multi-instance infrastructure, Bardie manages complex sales flows and real-time data synthesis natively — integrating with email protocols or chatbots, and architected from day one for final scaling on Gemini 2.5.
#2 Strategic Core
The core challenge was engineering a sales agent capable of executing ZMOT and Three Gaps methodologies across 7 distinct Buyer Personas with zero hallucinations and full contextual precision. This required building a RAG knowledge base that could handle complex, multi-persona conversations while maintaining strict response accuracy at every touchpoint.
The solution was architected around a proprietary PHP core (Instille Multi), ultra-fast inference via Groq, and a fully asynchronous messaging layer powered by SMTP and Cron Jobs. The result was a complete operational ecosystem including AI log dashboards, system mapping, and technical documentation shipped end-to-end in 7 days.
- Inference Engine & Scalability: Initial implementation with Groq (LPU) for minimum latency in the PoC phase, with a decoupled architecture ready for migration to Gemini 2.5.
- Proprietary Middleware (Instille Multi): Orchestrator built in PHP with a multi-instance architecture, enabling the deployment of multiple isolated environments (motorcycles, boats, etc.) under a single logical core.
- RAG Architecture (Retrieval-Augmented Generation): Dynamic knowledge base with hierarchical dependency injection to ensure precise responses grounded in 7 Buyer Persona archetypes.
- Async Communications Management: Integration of SMTP protocols and automation via Cron Jobs for managing sales threads and automated responses 24/7.
- Monitoring & Control (System Dashboard): Centralized control panel for supervising AI logs, SMTP server status, and real-time traceability of conversation flows.
- System Mapping & Technical Documentation: Fully documented architecture, including logical process mapping, decision flows, and implementation guides for product scaling.
- Frontend & Conversion Assets: Deployment of high-speed Landing Pages and end-user Dashboards focused on pre-sale metrics visualization and lead capture.
#3 Work Process
The Challenge: Can a production-ready AI sales agent capable of executing complex selling methodologies, be built and deployed in just 7 days? This project was a technical sprint to transform a generative AI concept into a fully functional pre-booking engine. Starting from zero, the entire stack was architected and shipped: from the RAG knowledge base and multi-tenant PHP middleware to async messaging infrastructure and real-time monitoring dashboards.
Every decision was made to eliminate the gap between sales psychology and technical automation deploying a system that thinks, responds, and converts 24/7 without human intervention.
- 01. Prompt Engineering & Sales Logic: Defined the business logic framework based on ZMOT and Three Gaps methodologies, modeling the RAG knowledge base for 7 Buyer Personas — zero hallucinations by design.
- 02. «Instille Multi» Architecture: Evolved the injector into a multi-tenant PHP infrastructure, enabling instant deployment of isolated instances with pre-configured cores under a single logical engine.
- 03. Async Implementation (PoC): Integrated Groq for ultra-fast LPU inference and built the messaging engine via SMTP and Cron Jobs, enabling autonomous management of sales flows 24/7.
- 04. Automatic RAG Optimization: Deployed a synthesis module that processes large-scale product datasets and automatically converts them into AI-optimized knowledge structures with hierarchical dependency injection.
- 05. Fine-Tuning & Gemini Scaling: Refined Bardie’s core through iterative optimization of sales processes and email threads. This phase consolidates the methodology to transfer the trained «brain» from the PoC into the final production infrastructure on Gemini 2.5, ensuring superior performance in the commercial release.
#4 Results
Bardie compressed what typically takes months into a 7-day delivery cycle — from blank canvas to a fully operational lead-capture system. The pre-booking engine runs autonomously, managing 3 to 4-email sales sequences with full contextual coherence and zero human intervention.
Built on a multi-tenant architecture, the platform is designed to scale across unlimited business instances, each with independent product catalogs and fully customizable sales logic. The PoC was validated with real client interactions email sequences delivered, responded to, and converted autonomously, confirming the system’s readiness for full commercial rollout. 🚀😊
More from Kerning LABS®
ClickConvert
PHP framework and ClickFlow® engine for high-performance web conversion.
Profe por horas
A flexible E-learning platform rethinking the price-per-hour model
Saldedudas
Real Estate App that finds the best buying options based on your financial health.