Engunity X : AI-Powered SaaS Platform for Research, Code Generation and Data Analysis

Abstract: Engineering students routinely juggle five or more disconnected tools in a single work session—a paper search engine, a chatbot, a code editor, a sandbox runtime, and some method for pressure-testing assumptions—and every context switch between them discards the mental state built up in the one just left. This paper presents Engunity X, a unified five-service SaaS platform that collapses that workflow into one shared-memory environment. The centrepiece is OmniRAG, a complexity-adaptive retrieval pipeline backed by a fine-tuned DistilBERT classifier (18,500 labelled queries, 91.3 % macro-F1) that routes each request to the most appropriate of four strategies: direct generation, hybrid dense-sparse retrieval, knowledge-graph traversal, or recursive chain-of-thought. On a 500-query benchmark drawn from publicly available CS documentation, OmniRAG reached 88.0 % retrieval accuracy on multi- hop questions—24 percentage points above a standard single-strategy FAISS baseline. A Docker-sandboxed Code Lab corrected 70.7 % of seeded program errors autonomously in one iteration. An adversarial Decision Vault flagged 78 % of logically weak arguments against a human-rated gold set. The full stack sustained P95 latency below 500 ms at 500 concurrent users on commodity hardware.

Keywords: retrieval-augmented generation; autonomous agents; knowledge graph; secure sandbox; adversarial reasoning; DistilBERT; engineering education; SaaS.