Pranjali Sonawane

Deploying a self-contained AI coding agent for on-premise inference on NVIDIA DGX hardware — owning the full deployment path from model serving (vLLM, OpenAI-compatible endpoints) through GPU diagnostics, process management, and configuration of locally hosted endpoints. Optimised inference for low-latency, data-sovereign use in air-gapped settings; instrumented monitoring hooks to evaluate deployed model performance and trigger adjustments. EDB Grant System Automation: Designed and built a multi-agent ML/LLM system to automate grant application intake, eligibility assessment, and documentation processing; owned end-to-end system design including architecture documentation, scalable folder structure, and safe tool-integration patterns. Implemented input/output guardrails, observability, and human-in-the-loop handoff to maintain model reliability and compliance — directly transferable to regulated financial-services workflows.

Represent Agora’s real-time engagement platform on campus; organise workshops on conversational AI, voice agents, and real-time multimodal ML applications.

Led a team of 5 engineers on a document-intelligence pipeline that lifts OCR accuracy across 100+ client invoices; designed and integrated APIs and backend services; evaluated and adopted new tooling, driving a 15% improvement in solution accuracy and 30% reduction in development and release timelines. Performed unit, integration, and end-to-end testing with Jest to ensure correctness and reliability; authored design and process documentation for technical and non-technical audiences. Translated ambiguous business requirements into technical solutions across cross-functional teams.