Experiments
01 Mar 2026
Clear Chain-of-Thought (ClearCoT)
How we implemented philosophical and logical thinking into the core of AI development to eliminate the 'Token Wall' and optimize SLM efficiency.
#LLM#Reasoning
→System_Log // V.2.0
Artifacts
Analyzing the intersection of logic, machine learning, and human cognition.
Experiments
28 Feb 2026
AI Infrastructure Planner
An interactive, highly accurate capacity planning toolkit for AI Engineers, DevOps, System Administrators and Researches. This tool mathematically calculates VRAM for Large Language Models (LLMs), accurately sizes RAM/Disk overhead for Vector Databases, and automatically compiles a complete suggestion of hardware for your workflow.
#Infrastructure#LLM
→
Philosophy
14 Feb 2026
The High Cost of Memory
Why is memory so expensive? While the market discusses RAM and SSD costs, a more subtle shift is occurring in how we deliver memory to digital devices—and what it costs us as humans.
#Philosophy#Human-Machine Interaction
→
Computer Vision
08 Feb 2026
HTR: Making Manuscripts Searchable
How we trained a custom machine learning model to read the handwriting of poet Milutin Bojić, unlocking full-text search for digital archives.
#Computer-Vision#eScriptorium
→
Heritage
08 Feb 2026
Post-OCR Correction: The Last Mile of Digitization
Raw OCR is not enough. We are building a parallel corpus to train models that clean up the 'noise' in our digital heritage.
#OCR-Correction#Datasets
→
Experiments
08 Feb 2026
Local AI Agents: 100% Offline with Gemma 3
Deploying autonomous Serbian AI agents on consumer laptops (8GB VRAM). A complete breakdown of infrastructure, the Agno framework, and fine-tuning Gemma 3.
#Local-AI#Agents
→
LLM
08 Feb 2026
Optimizing Whisper for Serbian
A case study in reducing Word Error Rate (WER) by ~85% for the Serbian language through targeted fine-tuning and hardware-aware architectural constraints.
#LLM#Fine-Tuning
→
Governance
18 Dec 2025
Safeguard-DH: Governing the 'Two Speeds' of AI
Our latest analysis reveals a friction in the cultural heritage sector: rapid personal AI adoption vs. stagnant institutional governance. Here is our protocol to fix it.
#AI-Ethics#ORRI
→
Heritage
15 Aug 2025
The Hidden Value of Old News: Why LLMs Need Cultural Heritage Data
The internet is not the sum of human knowledge. We digitized 50 years of local Newspapers archives. This approach can fill the information gap on internet that missing before 1990.
#LLM-Data#Digital-Humanities
→