Kutup
Kutup is developed by ASP Dijital as a structured ecosystem of specialized tools. The production engine, Kutup NQ, is a domain-constrained industrial assistant that combines curated datasets, retrieval pipelines, and guided model synthesis to answer technical questions with visible evidence.
Principles that guide every tool in the ecosystem.
Built incrementally: launch one reliable tool, validate it in production, then expand.
Neural Query is the first production app, focused on industrial automation, IT/OT, ML, protocols, and data-system fundamentals with evidence-aware output — intent routing, embeddings, and evidence-quality calibration run on-device in a 21 KB WebAssembly kernel.
NQ routes a query through custom datasets, browser-based ML signals (WebAssembly kernel: intent classifier, embeddings, quality calibration), and guided LLM synthesis, with selective Wikipedia/RSS enrichment when more context is needed.
Flow Maker turns plain-language process descriptions into visual flowcharts — on-device WebAssembly ML, deterministic, no LLM; Forge generates context-specific industrial mini-apps. Protocol LLMs (Modbus, OPC, PLC) are next.
One architecture and one UX language across every Kutup application.
NQ rejects out-of-scope prompts and only accepts supported English and Turkish industrial-domain requests.
Custom JSON datasets are curated and versioned, then expanded with bilingual retrieval paths to improve technical recall depth.
Session history, thinking states, tool-usage chips, and inspectable evidence panels keep the answer flow visible instead of opaque.
A managed LLM layer synthesizes answers from the evidence packet, while browser-side ML assists ranking and interaction quality on the frontend.