Blog
Articles on building AI agents, knowledge graphs and real cases from industry and medicine.
Articles
- 98.4% accuracy and ₽1bn of impact: how we found analogues in the SIBUR catalog — a case study of material analogue matching: 730,000+ analogue pairs, 400,000+ items, a trip to the plant and the mistakes along the way. Read: /blog/sourcematch-mtr-analogs
- AI PDLC: right diagnosis, wrong patient — a review of Sber's whitepaper: the corporate bottleneck is not the speed of writing code but the management processes around it. Read: /blog/ai-pdlc-diagnosis
- How to adopt AI in business without losing money: a step-by-step guide — how to find the right task through process diagnostics, calculate the effect honestly in FTEs and choose a vendor. Read: /blog/how-to-implement-ai-in-business
What we write about
- Adoption cases with numbers and with what went wrong — not just the final metric
- The effect calculation method: FTEs, applicability thresholds, evidence statuses
- The engineering side: agentic systems, meaning-based search, knowledge graphs, working with Russian standards
- Reviews of other people's material on AI adoption — with the claims checked, not retold
About Sandboxer
Sandboxer is a Russian AI company. 40+ AI projects for 7 enterprise clients in 2 years, >₽1bn confirmed impact. A portal of AI agents for a major petrochemical holding. A team of 20, no venture funding.