Rojan Upreti - AI and Software Engineer

Built a one-click installation platform for private AI inference and healthcare AI agents on NVIDIA GB10 hardware, running all models locally to meet on-prem/no-cloud requirements, using sandboxed OpenClaw/NemoClaw agents bounded by OpenShell to safely automate hospital operational workflows. Secured hospital AI-agent operations through MCP-based tool boundaries, human approval gates, and MongoDB-backed audit events, preventing direct exposure of EHR and ServiceNow credentials. Earned recognition from MongoDB and was featured by NVIDIA Healthcare AI for innovation in private, secure hospital AI infrastructure.

Built a full-stack compliance dashboard with AI-powered PHI detection across codebases and infrastructure, de-identification tooling using Safe Harbor and Expert Determination methods, and automated findings mapped to specific HIPAA citations and SOC 2 criteria. Designed an RBAC system with 7 clinical roles, 32 row-level security policies, audit logging, and an email-based invite flow. Engineered a PHI inventory system with scanner sync, risk scoring, retention enforcement, and one-click audit-package PDF generation.

Built a web and mobile platform for Passaic County serving 78,000+ SNAP recipients using Next.js, Node.js, TypeScript, and Express to automate work-hour compliance tracking. Architected a distributed, multi-tier RBAC system with ownership-validated APIs, JWT middleware, row-level database security, and audit logging. Added automated document intelligence, multi-stage approval workflows, and AI-assisted anomaly detection—projected to save Passaic County up to $2M annually.

Designed a middleware caching framework for RAG pipelines that decomposes compositional queries into sub-queries, enabling partial cache hits and reducing redundant LLM API calls by 58.1%. Built SHA-256 and pgvector HNSW indexing, cutting median latency by 47.4%, and used token-level entropy gating to achieve 94.1% hit precision. Released as an npm package and Python SDK.

Led design and development of a multi-platform HIPAA audit tool (web app, CLI, and VS Code extension) for real-time compliance analysis. Built with Azure OpenAI under a BAA, Firestore, and GitHub OAuth, with rule-based scoring, severity tagging, file-level evidence extraction, and compact JSON reports. Executed security audit workflows aligned with HIPAA technical safeguards.

Designing a HIPAA-compliant unified patient portal integrating Athena and Epic EHR systems via HL7 v2, FHIR, and SMART on FHIR APIs to aggregate and normalize 10+ categories of health data in real time. Implementing OAuth 2.0, JWT, RBAC, and AES-256/TLS encryption. Delivering a web and mobile platform for patients to access longitudinal records and for providers to view consolidated clinical data.

Developed an AI-powered interactive web platform that visualizes real exoplanet data and integrates a conversational AI assistant for astronomy education. Built with React and Node.js using NASA and Caltech datasets for 3D planetary exploration. Implemented OpenAI-powered story generation and Q&A, plus a scalable data pipeline, earning global nomination recognition at NASA Space Apps.

Designed a grocery-store POS system with role-based Cashier/Manager access and centralized database synchronization across 5 terminals. Implemented modular OOP architecture in C++ with real-time inventory updates, coupon/loyalty logic, and multi-method payments. Modeled the full architecture in UML and applied the software engineering lifecycle from requirements through validation.

Designed an integrated job application automation system combining a Chrome extension (JobFiller) with a cloud-backed web platform for real-time profile synchronization and ATS optimization. Built OpenAI-powered keyword extraction, Firebase Authentication and Firestore sync, secure document management, and REST APIs compatible with Chrome Manifest V3.