NQ · Retrieval
Versioned JSON collections, Turkish normalization, token-level BM25. Source-guided synthesis; evidence gaps are stated.
From field experienceto testable AI.
Kutup is ASP Dijital’s industrial AI workbench, focused on the engineering questions of the automation, energy and process teams ASP Otomasyon serves.
Before a bigger model, a better experiment.
01
Data
Units, provenance and versions. A clear line between synthetic examples and field data.
02
Method
Simple baselines, local computation and inspectable retrieval results.
03
Validation
Held-out evaluation, measurable acceptance criteria and engineering review.
What runs today.
Implemented behaviour is kept apart from research goals.
Flow Maker · Local parsing
Rule and classifier-assisted process parsing; editable graph, undo, portable exports.
Forge · Data checks
Bounded local parsing, explicit decimal and header controls, missing-value report, pairwise-complete correlation.
Lab · C/WebAssembly
Median, MAD, EWMA, Isolation Forest, regression, PID simulation, Cp/Cpk. Synthetic scenarios; anomaly inspection is not fault diagnosis.
Sentinel · Multivariate monitoring
Isolation Forest, PCA with Hotelling T² and SPE, EWMA; per-sensor contributions. Learns from normal operation, says “unusual”, never “fault”.
Spectra · Vibration analysis
FFT, Welch, envelope spectrum, spectral-kurtosis map, bearing fault frequencies. Synthetic records or your own data; the diagnostic decision stays with the engineer.
Horizon · Remaining life
Kalman local-linear-trend model (likelihood-tuned), Monte-Carlo threshold-crossing distribution; compared with linear and quadratic models.
The input to the next experiment.
Field accuracy needs authorized operational data, an independent test period and expert labels. The synthetic examples here are for inspecting methods.
Start with a production problem.
Let’s review your data source, your existing system and the decision you want to improve.