VedAI Clinical Psychiatry AI Assistant

AI Researcher (prior work, 2024 to 2025)

Prior work · 2024–2025 · Confidential

Designed reusable LLM instruction and context templates for psychiatric workflows, with a multi-agent human-in-the-loop review layer for sensitive clinical outputs.

Read my contribution ↓

Overview

VedAI was a private clinical-AI platform where I worked on structuring expert psychiatric workflows into reliable LLM instruction and context systems. The work focused on turning implicit clinical reasoning into reusable, well-documented templates and on building review paths that keep a human in the loop for sensitive outputs. This is prior work (late 2024 to mid 2025); details are shared at a high level under confidentiality.

My contribution

As an AI researcher on the project, I designed reusable instruction and context templates, a multi-agent review workflow with risk triage and escalation logic, and evaluation and documentation standards that aligned clinical and engineering review. I coordinated with clinicians and engineers on workflow refinement and UI/UX.

What I Worked On

  • Designed reusable LLM instruction and context templates covering 80+ psychiatric workflow scenarios, supporting more consistent outputs across complex cases.
  • Designed a multi-agent review workflow with risk triage and escalation logic to support human-in-the-loop review of sensitive clinical AI outputs.
  • Developed model evaluation and documentation standards for workflow refinement, output review, and review milestones.
  • Coordinated clinicians and engineers on workflow refinement, documentation, and UI/UX improvements.

Scope & Limitations

This was a contributor role on a private platform, and the work is not publicly shareable. The summary above describes my contributions at a high level rather than reproducing any proprietary content.

Technologies & Tools

LLM context designStructured prompt evaluationMulti-agent workflow designRisk triage and escalation logicMarkdown and JSON structuring

Deliverables & outcomes

  • Built reusable instruction and context templates across 80+ psychiatric workflow scenarios.
  • Standardized workflows across 12 psychiatric case categories, cutting new-scenario configuration time from 4 weeks to 2.
  • Established a multi-agent, human-in-the-loop review path with risk triage for sensitive outputs.
  • Created model evaluation and documentation standards that aligned clinical and engineering review.