Erim Evren Electrical & Computer Engineering · University of Washington · Seattle, WA

Builder · Systems · Seattle

Building AI systems, autonomous robotics, and real-world software under real constraints.

I am an Electrical & Computer Engineering student at UW focused on machine learning, LLM systems, computer vision, robotics, and reliability-critical software. I like building systems that move from idea to working product.

View projects Get in touch

About

End-to-end technical work

I am a full-time Electrical and Computer Engineering student at the University of Washington, admitted through the UW Academy program after 10th grade with a merit-based scholarship.

My work spans AI systems, real-time perception, autonomous robotics, distributed infrastructure, and reliability-critical C++. I focus on intelligent systems that use real-world context, respect hard constraints, and ship to users.

Experience

Where I have built

Software Engineering Intern GoDaddy June 2025 – September 2025 · Tempe, AZ

LLM-powered analysis of Amazon EMR production pipelines to detect upgrade risks across large dependency graphs.

  • RAG over pipeline code, configs, dependencies, and metadata for fast querying and reasoning.
  • Evaluation tooling for failure modes, inconsistencies, and risky upgrade paths.
  • Automated dependency analysis to cut manual work during production migrations.
  • Owned design through validation in a high-throughput production environment.
CubeSat Software Engineer Husky Satellite Lab December 2024 – Present · Seattle, WA

Reliability-critical C++ for CubeSat flight software.

  • Telemetry, serialization, and onboard data handling under strict memory and bandwidth limits.
  • Validation frameworks and failure tests for deterministic behavior where failure is not an option.
  • Communication and data paths where correctness drives mission success.
Autonomy & Computer Vision Engineer Washington Aerial Robotics June 2025 – Present · Seattle, WA

Real-time vision and autonomy for drones.

  • Perception pipelines from camera data to terrain representations for navigation.
  • Neural object detection with inference tuned for real-time performance.
  • Perception integrated with control for multi-drone coordination in dynamic environments.
  • Testing across lighting, motion, and real flight conditions.
Undergraduate Student Researcher ARC Lab, UW March 2026 – Present · Seattle, WA

GPS anti-spoofing and DOA research for drone navigation — Project ÆTHER / FusionFlight.

  • Synthetic GPS IQ via gps-sdr-sim with real RINEX ephemeris for a 5-element UCA at UW coordinates.
  • MUSIC for 2D DOA (power and carrier phase); carrier phase ~0.1° elevation error vs ground truth.
  • MVDR null steering (~40 dB suppression) in power and carrier phase variants.
  • CGA vs Euclidean benchmarks — optimized CGA MEX within 6.2% overhead of native C++.
  • Real RF from field tests (clean and jammed, 12.5 MHz). Stack: MATLAB, C++ MEX, garamon, Eigen, gps-sdr-sim.
Machine Learning Engineer Intern InventAI June 2026 – September 2026

Upcoming role: ML models and data pipelines for client-facing products.

  • End-to-end ownership from data processing and training through evaluation and deployment.
  • Focus on models that stay reliable in real production settings.

Projects

Selected builds

Expand any card for detail.

Project ÆTHER

Featured

GPS anti-spoofing · Signal processing · C++ / MATLAB

Full anti-spoofing pipeline: raw IQ from a 5-element circular array to azimuth/elevation of satellite sources, with null steering toward jamming.

  • MUSIC for 2D DOA in power and carrier phase variants.
  • Carrier phase MUSIC: ~0.1° elevation, ~0.6° azimuth vs known ground truth.
  • MVDR null steering ~40 dB suppression (power and carrier phase).
  • CGA C++ MEX benchmarked at 6.2% overhead vs Euclidean after optimization.
  • Validated on real recorded RF near UW (clean and jammed scenarios).
  • Synthetic pipeline: gps-sdr-sim, RINEX ephemeris, circular array phasing in MATLAB.
MATLABC++GPSDOAMUSICMVDRCGAAnti-spoofingDrones

BUZZE AI / BuzzBuilder

Product

AI product · Startup

LLM-driven content and workflows for schools, clubs, and small businesses.

  • Structured prompt pipelines and workflow tooling around LLMs.
  • End-to-end flows: prompting, backend logic, feedback loops, refinement.
  • Client-facing apps including poster generation from text prompts.
  • Explored facial recognition and design tooling for yearbook workflows.
  • Experiments with AI-assisted writing and education-focused GAN reuse.
LLMsPrompt engineeringStartupsContent generation

Transformer LM from scratch

ML

Machine learning · LLMs

Custom transformer in PyTorch to study attention, training dynamics, scaling, and reliability.

  • Attention, positional encoding, FFN, autoregressive decoding from first principles.
  • Training on large tokenized corpora; cross-entropy and perplexity evaluation.
  • Experiments on depth, context length, and width vs convergence and stability.
  • Tradeoffs among capacity, compute, stability, and generalization.
PyTorchTransformersLLMsDeep learning

LLM behavioral probing

Evaluation

AI safety · Model evaluation

How LLM outputs shift under prompt variation, adversarial inputs, and structured context.

  • Consistency and reliability across prompt perturbations.
  • Stress tests including jailbreak-style attacks.
  • Pipelines for variance, instability, and failure modes.
  • Analysis of input structure vs reasoning quality and intent alignment.
AI safetyRobustnessLLMsPrompt testing

LLM system evaluation & RAG

Systems

AI systems

RAG for structured reasoning over technical documents and codebases.

  • Retrieval design and its impact on downstream answer quality.
  • Hallucination and failure analysis in multi-step reasoning.
  • Tooling to compare prompts and retrieval configurations.
RAGEmbeddingsVector DBPython

Autonomous drone perception

Robotics

Computer vision · Real-time systems

Perception stack for Washington Aerial Robotics — camera to terrain for autonomous flight.

  • Pipelines from imagery to terrain representations for navigation.
  • Object detection with latency-aware inference.
  • Integration with control for multi-agent coordination.
  • Robustness across lighting, motion, and field flight.
Computer visionAutonomyObject detectionReal-time

Chess engine

C++

Algorithms · Game AI

Full C++ chess engine from scratch — bitboards, search, and UCI.

  • Bitboard representation for cache-friendly move generation.
  • Alpha-beta with TTs, null-move pruning, killer heuristic.
  • Polyglot opening book; full UCI; tested with Cutechess.
C++Alpha-betaBitboardsUCI
GitHub Repo

FlowScript

DSL

Developer tools · Language design

Diagramming DSL between Mermaid and UML — expressive without the ceremony.

  • Grammar and language spec from scratch.
  • Custom parser and AST in a TypeScript monorepo.
  • Web Component embed; SVG renderer.
TypeScriptParserSVGWeb Components

NCAA March Madness predictor

Data

Data science · Python

Python bracket system for the Kalshi challenge — scraping, alignment, simulation.

  • KenPom scraping plus ESPN API integration.
  • Normalization and fuzzy matching across sources.
  • Monte Carlo full-bracket simulation; Duke, Arizona, Michigan top contenders in final run.
PythonMonte CarloWeb scrapingSports analytics

Signal decomposition & music reconstruction

DSP

Signal processing · Math

Fourier-based decomposition of recordings and exploration of note/chord reconstructions.

  • Frequency decomposition via Fourier analysis.
  • Reconstruction experiments using notes and harmonics.
  • Links between representations and perceptual similarity.
FourierAudioPython

Low-latency order book

Systems

C++ · Matching engine

Deterministic event-driven matching engine exploring latency and throughput.

  • Event-driven limit order book in modern C++.
  • Cache-conscious layouts and structures.
  • p50/p99 latency and throughput instrumentation under load.
C++Low latencyData structures

Mermaid GPT

AI tools

Developer tools · LLMs

Natural language to Mermaid diagrams — prompts, parsing, structured output.

  • Text to Mermaid syntax with guardrails on structure.
  • Prompt and parsing pipeline for consistent diagrams.
LLMsMermaidPrompt engineering
GitHub Repo Live Site

Skills

Stack and depth

Languages

Python, C++, C, TypeScript, SQL, MATLAB

AI / ML

PyTorch, transformers, LLMs, RAG, vector databases, prompt engineering, model evaluation, deep learning, computer vision

Systems

Linux, Git, real-time systems, low-latency design, multithreading, MEX, performance profiling

Signal processing

GPS/GNSS, DOA estimation, MUSIC, null steering, CGA, Fourier analysis, array processing

Robotics / autonomy

Vision pipelines, autonomous navigation, multi-agent coordination, perception-to-control integration

Math

Linear algebra, probability, optimization, Fourier analysis, geometric algebra

Resume

PDF

Experience, research, and projects in a single file — use download for recruiters and applications.

Download PDF Open in browser

Contact

Let us build something

Research, internships, collaborations, or hard systems problems — I am easiest to reach by email.

Email eevren@uw.edu LinkedIn GitHub