~/kaustubh-lall
kaustubh@portfolio:~

whoami: Kaustubh Lall, Senior Machine Learning Engineer · GenAI & Agent Systems, San Diego, CA. I build production GenAI systems: conversational agents, RAG platforms, and the evaluation infrastructure behind them. I built ResMed's generative-AI assistant Dawn and the virtual-therapist agent Dusk, and my published medical-AI research is the proof I can ship in high-stakes domains.

  • San Diego
  • Los Angeles
  • New York City
View projects(20)Explore the map
$ ls ~/

press Ctrl K to search

~ $ ls ./featured-work

Featured work

A selection of production and research projects, spanning medical AI, agent systems, and open research.

status: active; proprietary; live demo availabledawn.md

Dawn

active

ResMed's AI-powered digital sleep-health assistant. I was one of the engineers who built it.

  • LLMs
  • Python
  • Cloud infrastructure
Read case study
status: private; proprietary; no live demodusk.md

Dusk

private

A virtual-therapist conversational agent supporting patients waiting to receive care, built with RAG, persistent memory, multi-agent architecture, and voice-to-voice interaction.

  • LangChain
  • RAG
  • Speech-to-text
  • Text-to-speech
Read case study
status: active; public code; live demo availablemmg.md

OSRS Market Tools (MMG)

active

Money-making-guide and Grand Exchange analysis for Old School RuneScape: deterministic margin analytics plus a leakage-checked ML research track on public market data.

  • Python
  • Pandas
  • OSRS Wiki API
Read case studyLive
status: active; public code; no live demodino-arena.md

Dino Arena

active

A reproducible, framework-independent simulation environment for fair agent benchmarking, not just a game.

  • TypeScript
  • Canvas
  • Headless simulation
Read case studyCode
status: research; public code; no live demoalgorithmic-solvability.md

Algorithmic Solvability

research

Controlled experiments on whether ML models detect that a task is governed by a compact algorithm.

  • Python
  • PyTorch
  • scikit-learn
Read case studyCode
status: active; public code; live demo availableneat-playground.md

NEAT Playground

active

NEAT neuroevolution implemented from scratch in C, evolving neural nets that play Flappy Bird, Snake, 2048, and Pac-Man, with interactive replay visualizers you can run in the browser.

  • C99
  • NEAT
  • SDL2
  • HTML/JS visualizers
Read case study

~ $ open ./constellation

Explore the project constellation

Every project as a node. Hover, click, or Tab through the graph.

Open the project constellation map
constellation.svg

20 projects mapped by category and relationship.

open the full map

~ $ cat experience.log

Experience

Download resume

  1. Feb 2024 – Present

    Machine learning and conversational AI for sleep health, from GenAI platform infrastructure to consumer products and clinical modeling.

    • Founding engineer on ResMed's internal GenAI platform: retrieval-augmented generation infrastructure, agent architectures (LangGraph tool-use agents), Terraform-managed AWS environments, security hardening, and LLM fine-tuning and evaluation pipelines reused across downstream products.
    • Built Dawn, ResMed's generative-AI sleep-health assistant, largely from the ground up: one backend powering both the myAir app and the web experience, and I built both surfaces.
    • Built Dusk, a virtual-therapist conversational agent for patients awaiting care: specialized agents, retrieval-augmented generation, persistent user-specific memory, voice-to-voice interaction, CBT-I-guided goals, and patient phenotype classification with adaptive guidance. IP filing pending.
    • Currently leading COMISA (comorbid insomnia and sleep apnea) modeling: clinical ML research on polysomnography data with interpretable baselines and clinician-in-the-loop label validation.
    • Heavy code-review and mentoring load across the GenAI platform: onboarding materials, contribution-process docs, and hundreds of pull requests authored and reviewed.

  2. 2021 – 2024

    Founding engineer at a health-tech startup running AI-backed, direct-to-consumer clinical trials.

  3. 2019 – 2021

    Computer vision for osteoarthritis diagnosis and treatment measurement with UCSD/VA radiology researchers.

  4. 2018 – 2020

    Machine learning for drug-transporter substrate classification (OAT1/OAT3).

~ $ ls ./publications

Publications

Independent research with the UCSD radiology group, alongside earlier biomedical machine-learning work.

~ $ whoami --verbose

About

I'm a machine learning engineer working in medical AI. My career has moved through three connected layers. It started in biomedical and medical-imaging research at UC San Diego: first a model that predicted drug-transporter binding without lab work (published in the Journal of Biological Chemistry), then computer vision at a radiology lab, measuring cartilage-disease treatment outcomes from MRI.

After a master's in machine learning at UCSD, I spent about four years as a founding ML and cloud engineer at Radicle Science, a health-tech startup running AI-backed direct-to-consumer clinical trials. I built reporting, data-normalization, and recommendation systems used across tens of thousands of study participants.

Today I build conversational AI and ML systems for sleep health at ResMed, including Dawn (ResMed's generative-AI sleep-health assistant) and Dusk (a virtual-therapist agent platform), and I keep publishing medical-imaging research through an independent collaboration with the UCSD radiology group. I've been based in San Diego, Los Angeles, and New York City along the way. The thread across all of it: take a real, often ambiguous question, build a testable model or system, and find out what breaks when it meets real users, data, latency, or clinical constraints.

Skills

aggregated from 20 projects · counts link to filtered views

Languages

ML & data

GenAI & agents

Web & product

Systems & infra

Games & simulation

Other