VIVECHANA × VIVEKA: A Two-Stage Architecture for Hallucination-Free AI Decision-Making in Multi-Domain Enterprise Systems
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Contemporary AI systems operating across enterprise knowledge domains exhibit two structurally distinct failure modes that existing architectures treat as one: (1) the wrong-problem failure, wherein a system reasons correctly but toward an incorrectly selected problem, and (2) the hallucination failure, wherein a system generates plausible but unverified domain conclusions. We present VIVECHANA × VIVEKA — a two-stage pre-computation architecture that eliminates both failure modes simultaneously and by design. VIVECHANA (विवेचना — discernment, pre-analysis) is a mathematical problem-selection gate that scores candidate problem spaces on three dimensions — Impact (I), Feasibility (F), and Leverage (L) — producing a scalar S = I × F × L × R prior to any knowledge architecture construction. The problem with the highest score is selected by computation, not interpretation. VIVEKA (विवेक — discriminative wisdom) is a pre-computed expert conclusion library wherein domain answers are stored as verified entries — not generated at query time. The combination achieves a provable guarantee: the right problem is selected before reasoning begins, and the domain answer is retrieved rather than generated, making hallucination architecturally impossible at the domain conclusion layer. We situate this work within the Dharmakīrti pramāṇa framework, demonstrate a first recorded deployment (AnnaPurna agricultural AI platform, 2026-03-26), and show how VIVECHANA × VIVEKA integrates with an existing 200+ service enterprise AI architecture via capability bitmask encoding and the Forja service declaration protocol.
Fields of Study & Subjects
Artificial Intelligence Applications
AI-based Problem Solving and Planning
Decision Support System Applications
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 26 March 2026
- Language
- EN
- Document type
- Preprint