Technical notes from our Pilot Builds.

Reference architectures, technology selection criteria, and rollout roadmaps — written for people who'll actually read the code and run the system, not company news.

AI & MLGPUVRAMHBM

How important are VRAM and HBM in AI infrastructure: from inference to fine-tuning

When choosing AI infrastructure, FLOPS usually gets mentioned first — but in practice, VRAM capacity and HBM bandwidth are often what determine which models can run, at what batch size and latency. How to size VRAM for production, why HBM differs from GDDR, and when you need multiple GPUs.

15/06/20267 min read

DevelopmentCybersecurityAI SecurityZero Trust

Security in the age of AI: new risks and zero-trust principles for production systems

AI agents and LLMs integrated into production systems open up a new attack surface — prompt injection, data poisoning, deepfakes — adding to, not replacing, traditional security risks. Zero-trust principles, secure SDLC, and a practical checklist for protecting AI-enabled systems.

14/06/20267 min read

IoT & SensorsLiDARPoint CloudCOPC

Reading part of a massive point cloud file: classification, octrees, and COPC for LAS/LAZ/PTS data

A LiDAR LAS/LAZ scan of a plant or a road corridor can weigh dozens of GB and contain billions of points. Most real tasks — fetching only ground points, only one structure in an area, or a low-resolution preview — don't need the whole file. How LAS/LAZ/PTS organize classification, bounding boxes, and octrees, and how PDAL, Entwine, and COPC exploit them to load only the points you need — cutting system load and speeding up large-scale 3D processing.

13/06/20268 min read

Data AnalyticsdbtAirflowELT

Building an ELT pipeline with dbt and Airflow: from zero to a production dashboard

A concrete roadmap from scattered data across multiple systems to a data warehouse you can trust — with dbt for transforms, Airflow for orchestration, and data contracts so teams don't break each other's dashboards.

12/06/20268 min read

Data AnalyticsOCRData QualityMaster Data Management

Data digitization: from paper, Excel, and legacy systems to a single source of truth

Most data digitization projects fail not because OCR misreads characters, but because no one reconciles, normalizes, and merges the data afterward. A digitization pipeline architecture: OCR/ICR, validation, entity resolution/MDM, and an incremental migration strategy — why "one source of truth" matters more than "99% character accuracy".

11/06/20268 min read

AI & MLQwen3-VLComputer VisionOCR

Deploying Qwen3-VL for computer vision in production: from OCR to visual inspection

Qwen's latest Vision-Language Model unlocks a class of computer vision problems that traditional OCR and CV pipelines struggle with — but production deployment requires a different mindset than calling a demo API. Use-case selection criteria, model sizing, and how we integrate it into Pilot Builds at KonexForge.

10/06/20267 min read

DevelopmentSEOAI SearchStructured Data

Optimizing a website for SEO and AI search: when ChatGPT, Claude, and Perplexity read your pages too

Alongside Googlebot, crawlers from ChatGPT, Claude, Perplexity, and Google AI Overview now read website content directly to synthesize answers — not to rank pages. Concrete technical changes: robots.txt for AI crawlers, JSON-LD structured data, writing 'quotable' content, and why Core Web Vitals still matter.

09/06/20268 min read

Server & DatabaseTimescaleDBClickHouseIoT

Database architecture for IoT systems: when to use TimescaleDB, when to use ClickHouse

The two most popular time-series engines for sensor data — selection criteria, schema design, and a data lifecycle strategy so costs don't explode with device count.

08/06/20267 min read