Blog

Technical articles, no opinion pieces.

We write about what we do: Lean Six Sigma, industrial Machine Learning, agnostic architectures. Articles are dense, quantified, and aimed at the practitioner who wants to understand — not the decision maker who wants to feel reassured.

  1. 06

    Conformal Prediction: Guaranteed Confidence Intervals for Industrial ML

    Your XGBoost predicts a scrap rate, but with what uncertainty? Conformal prediction yields intervals with a proven coverage guarantee, no distributional assumptions.

    17 min read conformal-predictionuncertaintyprediction-intervals
  2. 05

    Recursive Embedding Fine-Tuning: Learning Without Forgetting (the Funes Paradox in Production)

    How Nika OS continuously retrains its embedding model with LoRA, measures the real gain (Recall@k, MRR, nDCG), and guards against catastrophic forgetting. Real curves, empirical gate, nothing fabricated.

    18 min read embeddingsfine-tuningrecursive-fine-tuning
  3. 04

    Holding, Parent-Subsidiary Regime and Tax Consolidation: Structuring Your Assets in 2026

    Setting up a SAS or SARL holding company, parent-subsidiary regime (articles 145 and 216 of the French Tax Code, 95% exemption on dividends received).

    19 min read holdingtaxationparent-subsidiary regime
  4. 03

    The real cost of a RAG stack in production — SMB numbers

    Honest breakdown of a production RAG stack: Anthropic/OpenAI tokens, Qdrant cloud vs self-host, compute infra, dev maintenance.

    12 min read ragcosttco
  5. 02

    When one agent isn't enough — orchestrator pattern for SMBs

    Limits of a single LLM agent (context, bias, rate limit), concrete thresholds for switching to multi-agent, patterns.

    11 min read agenticaimulti-agent
  6. 01

    Can an agentic system replace an ERP?

    3-year TCO of a classic ERP vs an agentic stack (LLM + RAG + LangGraph + DB) for a 15-50 employee SMB.

    12 min read agenticaimanufacturing