Back to Home

Niloy Roy

Machine Learning Engineer

I build intelligent systems and dynamic web applications.

About

Machine Learning Engineer specializing in agentic AI systems that run in production. I build conversational assistants, multi-agent products, and clinical-research platforms from architecture through deployment. My work spans persistent agent memory, vector search and reranking, ML pipelines, observability, and modern web product development.

Experience

The Analytical Engineer

Logic, analysis, and data-driven systems. This is where I build the foundation, from complex ML models to scalable infrastructure.

  1. DEC 2025 — PRESENT

    Built a client-facing conversational assistant with LangGraph and MCP, plus ML classifiers, persistent pgvector memory, Redis state management, and Grafana monitoring for production agent workflows.

    • LangGraph
    • MCP
    • PostgreSQL
    • pgvector
    • Redis
    • Grafana
    • ML Pipelines
  2. Sept 2025 — PRESENT

    Founder & Lead Developer · Dishpit - Restaurant Automation with AI

    Architected and developed "Dishpit," an AI-powered restaurant management web app using React, Vite, and Supabase. Engineered a suite of context-aware LLM assistants for real-time suggestions on scheduling, inventory, and menu optimization.

    • React
    • Vite
    • TypeScript
    • Supabase
    • Tailwind CSS
    • shadcn/ui
    • LLMs
  3. SEP 2024 — PRESENT

    Machine Learning Engineer (Research) · St. Michael's Hospital - Unity Health

    Engineered the Meta-Analysis Assistant (MAA), a LangGraph agent for data extraction and statistical computation in R. Built the STACER Agentic System with virtual-patient and automated-rater agents for psychiatry training, then deployed the applications on AWS.

    • Python
    • LangGraph
    • LLMs
    • R
    • AWS
    • React

The Creative Human

Creativity, intuition, and high-speed human interaction. This is where I connect, adapt, and build rapport in fast-paced, real-world environments.

  1. Aug 2022 — Present

    Crafted premium cocktails for high-profile events at venues including Rogers Centre and Scotiabank Arena. Managed high-volume service during live games and concerts.

    • High-Volume Service
    • Mixology
    • POS
    • Event Service
  2. Jun 2024 — Jul 2025

    Crafted a diverse range of classic and innovative cocktails in a fine dining environment, provided exceptional customer service, and processed transactions using Silverware POS.

    • Mixology
    • Customer Service
    • POS
    • Silverware
  3. Aug 2024 — Dec 2024

    Managed bar inventory and restocking, supported bartenders during high-volume periods (200+ customers/night), and prepared garnishes and mixers.

    • Inventory Management
    • High-Volume Support
    • Food Prep

Projects

  1. Extended the STACER Agentic System with an embodied virtual patient, streaming audio tokens through MuseTalk to generate realistic 2D facial expressions and lip-sync synchronized with the patient agent.

    • Python
    • MuseTalk
    • LangGraph
    • Agentic AI
  2. Dishpit - Restaurant Automation with AI

    Architected and developed 'Dishpit,' an AI-powered restaurant management web app using React, Vite, and Supabase. Engineered a suite of context-aware LLM assistants for real-time suggestions on scheduling, inventory, and menu optimization.

    • React
    • Vite
    • TypeScript
    • Supabase
    • Tailwind CSS
    • shadcn/ui
    • LLMs
  3. Meta-Analysis Assistant (MAA)

    Developed an LLM-powered agent built with LangGraph to automate data extraction, statistical computation using R as a core engine, and report generation for systematic reviews in psychiatric research.

    • Python
    • LangGraph
    • LLMs
    • R
    • Agentic AI
  4. STACER Agentic System (SAS)

    Built an LLM-powered simulation platform for psychiatry residency training featuring a virtual patient agent and an automated rater agent that delivers STACER-aligned feedback.

    • Python
    • LLMs
    • AI
    • React
    • LangGraph

Open Source Contributions

TherapeuticNLP

A multi-dimensional Natural Language Processing (NLP) framework for analyzing and assessing the quality of therapeutic conversations

Tools

Python, Matplotlib, NumPy, Pandas, Scikit-learn, SMOTE libraries

Package Details

Maintainers: Roy Niloy, Sarmed Shaya, Dylan Weston and Zachary Cote

License: MIT License

Requires: Python >=3.6

Classifiers

  • Intended Audience: Science/Research
  • License: OSI Approved :: MIT License
  • Operating System: OS Independent
  • Programming Language: Python :: 3
  • Topic: Scientific/Engineering :: Artificial Intelligence

Certifications

  1. 2025
  2. 2025

Publications

ORCID: 0009-0009-3199-6983 ↗
  1. 2026

    Enhancing Psychiatry Training Using an Agentic AI Simulated Consultation Tool: Prospective Cohort Study

    JMIR Medical Education 12, e88580

  2. 2026

    Subgenual anterior cingulate cortex in major depression: a systematic review and meta-analysis of MRI studies with patients and healthy controls

    Translational Psychiatry 16, 225

View My Resume