edwinkim / README.md
Preview

Hi there, I'm Edwin 👋

I'm a freshly graduated student at UC Berkeley for Computer Science. I specialize in AI and have a passion for building tools that solve complex real-world problems at scale.

Contact Info

class ContactInfo:
    email = "edwinkim0509@berkeley.edu"
    phone = "(310) 617 6693"
    linkedin = "linkedin.com/in/kedwin"
    location = "Berkeley, CA"
    education = "UC Berkeley, CS (2025)"

Current Focus

  • Creating projects that I would use in my personal life.
  • Contributing to real-time open source projects.
  • Taking care of my kitten.

Pinned

Public

Hermes

Grid Outage Prediction & Restoration Time Estimator. 94% precision in forecasting grid failures 48 hours in advance. Real-time geospatial dashboard processing 10M+ daily data points.

Random Forest
Geospatial
Python
Python
12
4
Public

CALI

Real-time Stress-Tracking Platform. Custom BiLSTM neural network processing 5 heterogeneous streams. Converges 40% faster than baseline.

BiLSTM
Signal Processing
Real-time
Python
8
2

Technical Skills

Skill Proficiency (Contribution Style)

Less
More

ML / AI

PythonTensorFlowPyTorchBiLSTMRandom ForestSignal ProcessingTime Series

Backend

API DesignDatabase OptDist. SystemsCachingFastAPIDjango

Frontend

ReactNext.jsTailwind CSSData VizReal-time UI

Data Eng

ETL PipelinesData FusionGeospatialPandas/Numpy

Infrastructure

GitDockerAWSLinuxCI/CD

Contribution Activity

ML Engineer

CALI
2025
  • Architected real-time stress-tracking platform using custom BiLSTM networks.
  • Fused 5 heterogeneous time-series streams with <10ms tolerance.
  • Generated 100,000+ synthetic data points for model training.
  • Achieved 40% improvement in model convergence speed.

SWE Intern

MyFitnessPal
2024
  • Refactored core food-log query engine serving 50M+ DAU.
  • Reduced request latency by 25% and API payload size by 40%.
  • Optimized queries for millions of concurrent sessions.

Computational Researcher

UC Berkeley
2023
  • Designed scalable materials informatics pipeline for 1,000+ samples.
  • Automated extraction of 50+ high-dimensional features.
  • Created dataset of 50,000+ data points, improving accuracy by 23%.