Sohen Patel

Built an enterprise Ontario Health News Agent in Microsoft Copilot Studio end-to-end, deployed to 5,000+ employees org-wide Created a 259-item human-verified gold-standard dataset for 3 healthcare tasks; built an end-to-end LLM evaluation framework for 5 models (GPT-5, Claude, Gemini, MS Copilot, Perplexity) across 13+ metrics using RAGAS, DeepEval, and LLM-as-a-Judge Conducted systematic prompt-engineering experiments across 4 strategies, finding Chain-of-Thought prompting on a 5-year dataset increased agent F1 score from 0.25 to 0.42 (+68%) and reduced hallucination rate by 51% (78% to 27%); logged runs to Langfuse.

AI Research Intern - Ontario Health - Toronto, ON, Canada

(2026-07 - 2026-09)

• Built an enterprise Ontario Health News Agent in Microsoft Copilot Studio end-to-end, deployed to 5,000+ employees org-wide
• Created a 259-item human-verified gold-standard dataset for 3 healthcare tasks; built an end-to-end LLM evaluation framework for 5 models (GPT-5, Claude, Gemini, MS Copilot, Perplexity) across 13+ metrics using RAGAS, DeepEval, and LLM-as-a-Judge
• Conducted systematic prompt-engineering experiments across 4 strategies, finding Chain-of-Thought prompting on a 5-year dataset increased agent F1 score from 0.25 to 0.42 (+68%) and reduced hallucination rate by 51% (78% to 27%); logged runs to Langfuse.

AI Engineering Project Team Lead - Deloitte (Sponsored Practicum) - Toronto, ON, Canada

(2026-05 - 2026-06)

• Built a 6-agent system that turns a client brief into consulting deliverables (WBS, RACI, RAID, weekly status report) with a human-in-the-loop revision loop, plus Deloitte template sync for DOCX/XLSX exports, backed by 129 automated tests
• Added an LLM routing layer with fallback across 4 backends (Claude, GPT, Gemini, DeepSeek) and prompt-cache warm-up to cut repeated token cost on agent calls
• Built a 13-rule Quality Gates engine (graph-based WBS cycle detection, RAID/RACI cross-artifact validation)

Graduate Engineer Trainee - ArcelorMittal Nippon Steel India (AM/NS) - Surat, India

(2024-08 - 2025-08)

• Used SAP datasets and project logs for real-time monitoring of compliance metrics and operational workflows, generating actionable insights that supported executive-level decision-making across 12 cross-functional departments
• Designed interactive Power BI dashboards to track project KPIs, resource allocation, and progress trends for data-driven planning across a large-scale plant expansion project from 7 to 14 million metric tons per annum (MMTPA)
• Identified workflow bottlenecks and execution gaps, contributing to a 16.7% reduction in project timelines

Master of Engineering - Emphasis: Data Analytics and Machine Learning - University of Toronto (2025-09 - 2027-01)

Bachelor of Engineering - Mechanical Engineering - L.D. College of Engineering, Gujarat Technological University (2021-07 - 2024-07)