J.A.R.V.I.S - Just Actionable Real-Time Visibility Intelligent System

J.A.R.V.I.S. is an AI assistant for incident triage in banking operations, developed as a capstone project in collaboration with HCL Technologies. When a production incident occurs, an operator describes the problem in a chat interface, optionally attaching a screenshot, log extract, or document, and receives a streamed answer that identifies probable root causes, affected components, and recommended next actions. Every answer is grounded in evidence: each claim is traceable to a specific runbook or past incident, and when no sufficiently relevant source exists, the system says so rather than guessing. J.A.R.V.I.S. is strictly decision support; humans retain full control, and no automated action is ever taken against a banking system.



Architecturally, the system comprises a React web application, a Java platform service managing accounts, documents, and access control, and a Python agent service that orchestrates the reasoning loop, retrieves knowledge through lexical search, and validates every response before delivery. A dedicated validation layer verifies citations, scores confidence, checks each sentence against retrieved evidence, and defends against prompt injection, while safety-critical controls are enforced by deterministic code rather than the language model. Attachments are redacted before use, and audit records preserve accountability while protecting personal data.