Stanford EE · Class of 2029

Diego Sanchez Tirado

Computer architecture and software. Seeking Summer 2027 internships in hardware, memory systems, and AI/SWE.

01 About

I’m Diego Sanchez Tirado, a Stanford electrical engineering student (B.S. expected May 2029). I’m leaning toward VLSI, chip manufacturing, and circuit design. The research I do is in computer architecture and memory systems, and I build software alongside it.

I’m a researcher in Stanford’s Engineering Robust Systems Group with Prof. Subhasish Mitra, and a developer at Verdnt Group. In summer 2026 I interned at Amazon Lab126 as a software engineer and as a humanoid-robotics apprentice. I previously shipped Trus v2, a generative interface that is now retired.

02 Work

Area-scalable memory

Research · Jan 2026 – present

Python models of LLM-inference throughput, latency, and memory under competing chip memory organizations. The work is with Stanford’s Engineering Robust Systems Group (Prof. Subhasish Mitra).

The models relate capacity, bandwidth density, and mesh or network-on-chip topology to batch size and context length. They compare a mesh-distributed, area-scalable backend with HBM under tensor- and pipeline-parallel plans, and report compute, memory, and interconnect in bandwidth-bound long-context KV-cache regimes.

Python

Trus

Product · shipped v2

A public generative interface. You describe an application in a sentence. Trus returns a tool built for that description, drawn from a trusted component library rather than from model-written UI code.

Shipped as v2 and later retired. A FastAPI service asks a model for a small JSON configuration. A Next.js canvas renders it with a fixed set of parts, on the order of 30. An event-sourced SQLite store sits under the generation path.

Next.js, FastAPI, Python, SQLite

FStea

Verdnt Group · April 2026 – present

Collectors and a cockpit for a refrigerated ready-to-drink tea brand. Python collectors bring in public launch signals. A Node cockpit is the working surface.

Module 7, a news sweep, verified at 102 launch rows from 48 articles across 8 chains, mean quality 0.90. Modules 7–11 feed Module 4 scoring. When the collectors have no data, the fused score matches the previous score exactly.

Python, Node.js

Plate Scanner

Personal

A phone web app for logging a Stanford dining-hall plate. A top photo, then a side photo. The portion comes from the plate’s dimensions and the two camera angles.

The browser checks the frame (whole rim, blur, glare). Grok vision names each food. A lookup against USDA FoodData Central, a restaurant page, or a brand label fills in closer numbers. A row is marked as a menu match only when it really is that dish. Otherwise it stays an estimate. The client is HTML, CSS, and JavaScript, served by a Python standard-library server.

HTML, CSS, JavaScript, Python

D.O.S.

Personal system

A private operating system for school, work, and the internship search. Domain agents keep their own context. A chief-of-staff agent handles the day, including calendar, mail, and a morning digest, and does not override the others.

Velvet, the console, generates the working surface. A deterministic ledger draws the same projections, so the data is still on screen if generation fails.

TypeScript

Hardware builds

Independent · Aug 2022 – present

Independent hardware, ongoing since 2022. A typing-force piezoelectric energy harvester with rectification and storage, measured for voltage and power against force and geometry.

A split ergonomic keyboard: CAD, a 3D-printed enclosure, and a custom PCB. A free-air active noise-control system with a microphone array and LiDAR on embedded hardware.

Onshape, custom PCB, 3D printing, embedded hardware

03 Experience

Developer

April 2026 – present

Verdnt Group

Verdnt Group builds custom AI systems for small and mid-sized businesses.
Building FStea: Python collectors and a Node cockpit. Module 7 verified at 102 launch rows, 48 articles, 8 chains, mean quality 0.90. Modules 7–11 fused into Module 4 scoring.

Researcher

Jan 2026 – present

Stanford Engineering Robust Systems Group · Prof. Subhasish Mitra

Area-scalable memory analysis for LLM inference. Python models of throughput, latency, and memory under competing chip memory organizations, relating capacity, bandwidth density, and mesh or network-on-chip topology to batch size and context length.
Comparing mesh-distributed (area-scalable) and HBM backends with tensor- and pipeline-parallel planning. Reports break out compute, memory, and interconnect in bandwidth-bound long-context KV-cache regimes.

Software engineering intern

June 2026 – Sept 2026

Amazon Lab126 · RAVEN, then Governed Knowledge Engine

Built a daily idempotent ingest of Outlook, Slack (direct messages and channels), and Confluence into a searchable Markdown corpus.
Implemented an LLM-as-judge classifier with a hybrid preset and an auto-discovered taxonomy, typed links, and a conflict queue. It reached at least 80% accuracy on a labeled eval and indexed more than 2,000 items on the first production run.
Re-architected the pipeline as a Governed Knowledge Engine. MindCraft is the retrieval source of truth. Prompt-injection and PII checks gate admission before anything is indexed for retrieval agents.

Humanoid robotics apprentice

June 2026 – Sept 2026

Amazon Lab126, Product Integrity / Reliability Lab · Dexmate “Labot”

Ran VR-headset teleoperation demos on the Dexmate bimanual humanoid, under a senior test automation engineer, and labeled datasets for manipulation-policy training.
Trained and evaluated ACT and diffusion imitation policies on physical bimanual tasks, then applied residual RL (TD3) beyond the demonstrated behavior.

Head GTM intern

July 2025 – Oct 2025

Code Four (YC X25)

Built go-to-market automation in n8n, Apollo, and HubSpot, with agents and scrapers covering more than 10,000 police contacts and email and call sequencing.
Reached more than 1,100 departments, yielding more than 70 qualified meetings, more than 10 MOUs, and more than 520 officers.

Cyber threat intelligence research

Mar 2024 – July 2025

Duke Pratt & Digi Americas · Prof. Arturo F. Ehuan

Co-authored LATAM Financial Threat Landscape 2025 with DigiAmericas and Recorded Future (LockBit, MisPadu, CL0P, Blind Eagle, HoraBot).
Maintained a Git-tracked indicator and TTP pipeline mapped to MITRE ATT&CK.

04 Education

Stanford University

Sept 2025 – May 2029

B.S. Electrical Engineering, expected May 2029

Coursework: Intro to Making (EE 40M), Vector Calculus for Engineers (CME 100), Programming Abstractions (CS 106B), Probability (CS 109), ODEs for Engineers (CME 102).

Autumn 2026: EE 101A, EE 100, CS 329Z, ARCHLGY 151.

Earlier, at NCSSM (North Carolina School of Science and Mathematics): CS4270 Fundamentals of Object Oriented Design, and CS4330/CS4460 Server-Side Development.

FTC RoboKnights 8569: 5th of 390 at the NC State Championship.

05 Skills

Hardware: Circuit design, PCB, soldering, CNC, 3D printing, laser cutting, piezoelectric instrumentation, microphone arrays, LiDAR.
Architecture: HBM, network-on-chip, tensor and pipeline parallelism. Models of capacity, bandwidth density, and KV-cache regimes.
Languages: Python, C++, JavaScript, TypeScript, HTML, CSS. English and Spanish.
ML / systems: LLM-as-judge pipelines, retrieval for agents, prompt-injection and PII gates, ACT, diffusion policies, TD3, FastAPI, Next.js, React.
CAD / tools: Onshape (certified), SolidWorks (FEA), Git, n8n. PLTW IED. MIT BSWI Python and GitHub.

06 Contact

Email: dsanchezt22@gmail.com
Phone: (984) 895-6386
GitHub: github.com/dsanchezt22
LinkedIn: linkedin.com/in/diego-sanchez-tirado
Resume: Download PDF
Location: Stanford, CA