#Challenge
The client needed to modernize genetic selection processes at a pedigree poultry farm with over 200 years of agricultural heritage and 70 years of breeding experience. Standard RAG architectures were not suited for genetic data — the system needed to reconstruct 5-generation genealogical trees, calculate inbreeding coefficients and rank breeding candidates using the BLUP (Best Linear Unbiased Prediction) algorithm. The challenge was building a production AI system that could process complex relational genetic data in near-real-time, integrate with existing .NET/Blazor applications, and deliver explainable decisions auditable by IFS Food, BRCGS and GlobalG.A.P. certification schemes. Additionally, Grupa Insight designed and built the public website rszew.pl on WordPress — presenting the farm's offer, breed catalog and contact information.
#Implementation
We designed and built a production-grade AI ERP system integrated with two existing .NET/Blazor applications via REST API and message queues. Instead of standard document retrieval, we built a custom pipeline that reconstructs genealogical trees, calculates inbreeding coefficients and ranks breeding candidates using the BLUP algorithm.
The system processes data through 8 pipeline layers: data ingestion, chunking, Temporal Tables for full history tracking, deterministic BLUP/EBV calculations, multi-criteria ranking, Explainable AI (XAI), anomaly detection, and a feedback loop for continuous adaptation.
Tech stack: Laravel 11 / PHP 8.3, Vue.js 3 + Inertia.js, MS SQL Server 2022 with Temporal Tables and Columnstore Indexes, Redis 7, Kubernetes + Docker, MLOps pipeline.
To complement the AI pipeline, we also designed the data collection layer: a portable egg weighing system for field workers using KERN PCB scales with BLE adapters, allowing real-time egg weight recording per individual hen (hen_id) directly at the nest. This eliminated manual data entry and ensured that phenotypic data entering the BLUP pipeline is accurate, timestamped and traceable to individual birds.
In parallel, we designed and implemented rszew.pl — a WordPress website presenting the farm's breed catalog, Rosa breed lines, day-old chick offer and parent sets, optimized for performance and SEO.`
#Results
Pedigree analysis time reduced from weeks to minutes. Annual genetic progress improvement of 3-8% versus classical selection methods. Reduction of maintained breeders by 10-25% through more precise candidate ranking. Full auditability of AI decisions for IFS Food, BRCGS and GlobalG.A.P. certification schemes — explainability was a certification requirement, not a nice-to-have. Real-time egg weight data collection per individual hen, eliminating manual entry errors and improving phenotypic data quality entering the BLUP pipeline. The system continues to evolve through its feedback loop, improving ranking accuracy with each breeding cycle.