Skip to content

diego@stanford:~$ whoami

Diego Sanchez

I build ML systems that ship.

Adaptive, reliable ML systems — built for uncertainty.

edu: Stanford University — B.S. Computer Science (AI) & Statistics · Class of 2028

now: Research Engineer Intern @ Coinbase, ML Platform

prev: Microsoft (M365 Copilot Search NLP) · Stanford Medicine (LLM research)

$ request_resume$ view_workgithub ↗linkedin ↗

resume available on request — opens your email app

Diego Sanchez

diego@stanford:~$ cat ~/research/rollout-promotion.md

Current research

Research Engineer Intern, ML Platform · Coinbase · Jun 2026 – Present

problem:

Many model rollouts rely on predefined traffic stages and manual promotion decisions.

framing:

My current work explores a more adaptive framing: treating rollout promotion as sequential decision-making under uncertainty.

method:

The approach models the relationship between promotion decisions and observed evaluation signals with a Gaussian process surrogate, so every decision carries an explicit estimate of its own uncertainty.

Acquisition functions from Bayesian optimization weigh the value of promoting further against what is still unknown.

Safety constraints and rollback gates bound the decision space: at each step, a rollout can hold, advance, or roll back.

evaluation:

The methodology is evaluated offline, through counterfactual analysis.

note: this section presents a public-safe overview of the methodology and excludes internal implementation details, operational data, and results.

$ open experience/coinbase.md

diego@stanford:~$ ls ./experience

Shipped systems

Microsoft

Software Engineering Intern, M365 Copilot Search NLP

Jun 2025 – Sep 2025 · Redmond, WA

An LLM-powered triage pipeline for M365 Copilot Search feedback — guardrails, human-in-the-loop review, and PII removal at log scale.

model quality — evaluated on 1k+ queries

recall — evaluated on 1k+ queries
97%
precision — evaluated on 1k+ queries
92%

privacy — across 50k+ logs

PII removal — across 50k+ logs
99%+
estimated cost — by automating triage workflowsestimate
~$430k → ~$4.2k/yr

figures from Microsoft internship, Jun 2025 – Sep 2025

$ open experience/microsoft.md

SURF Stanford Medicine

LLM Developer Research Assistant

Jan 2025 – Jun 2025 · Stanford, CA

hospital data → secure backend → analytics pipeline → user interface

A HIPAA-compliant FastAPI backend and LLM for real-time unit analytics and predictive forecasting, with automated test suites benchmarking outputs against reference responses.

HIPAA-conscious backendautomated evaluationreference-response testing

95% reported output accuracy

methodology details pending: task, dataset, split, sample size, and metric definition

$ open experience/surf-stanford-medicine.md

$ cat ./education

Stanford University logo

Stanford University — B.S. Computer Science (AI Track) & Statistics · Class of 2028

coursework: Data Structures & Algorithms, Computer Organization & Systems, Probability, Linear Algebra

clubs: Stanford Computer Forum, SOLE, ACM, Stanford AI Club

also: Claude Ambassador Builder @ Anthropic (Jan 2026 – Present) · UNBOXED Fellow @ Jane Street (Jul 2024 – Aug 2024)

$ open experience/programs.md

diego@stanford:~$ ls ./projects

Selected systems

InferScale

ML PlatformBackendAI Systems

~/projects/inferscale

system:

FastAPI inference microservices with Docker Compose, multi-mode routing, benchmarking harness, and C++17 deterministic replay.

idea:

Implemented routing for speculative and disaggregated prefill/decode modes with benchmarking harness.

capability:

Added C++17 deterministic replay on traces for reproducible performance analysis.

PythonFastAPIPyTorchDocker ComposePrometheusC++17
View repository ↗$ open projects/inferscale/README.md

Hospital LLM

Healthcare AINLP SystemsBackend

~/projects/hospital-llm

system:

HIPAA-conscious LLM backend for querying hospital operations data through schema-aware tools.

idea:

Built tool-calling backend for structured hospital data queries with schema grounding.

capability:

Created evaluation strategy for model accuracy and consistency on clinical operations data.

PythonLangChainDockerSQLFastAPI
View repository ↗$ open projects/hospital-llm/README.md

BUICU

Research EngineeringHealthcare AI

~/projects/buicu

system:

Bayesian ICU forecasting model using conjugate inference with an interactive Streamlit interface.

idea:

Implemented Monte Carlo simulation for uncertainty quantification.

capability:

Deployed interactive Streamlit dashboard for clinical analytics exploration.

PythonNumPySciPyStreamlitBayesian Inference
View repository ↗$ open projects/buicu/README.md

$ ls ./projects --all

  • StudyBuddyConnect

    Platform connecting students for collaborative learning and study session coordination.

    [GitHub]

  • HeadStart

    Project management and productivity tool for teams and individuals tracking goals.

    [GitHub]

  • Trade Bot

    Automated trading bot with algorithmic strategies, risk management, and performance analytics.

    [GitHub]

  • Chromatic Tuner

    Web-based chromatic instrument tuner with real-time audio processing and visual feedback.

    [GitHub]
$ view_all_on_github

diego@stanford:~$ tree ~

How I think

The same content at full depth — experience docs, project READMEs, and research focus in a navigable virtual filesystem. Browse it like a repo: ls, cat, grep.

15 documents · document search · tab completion · keyboard navigation · deep links

Enter Engineer Mode →

diego@stanford:~$ contact

Contact

new_message

diego@stanford:~$ mail dsanh14@stanford.edu

email: dsanh14@stanford.edu

Opens in your mail app — or copy the address above.

githublinkedinresume (request by email)