K.P.
K.P.

Kevin Pyo

scroll

Data science · UC San Diego

Turning random messy real world data into systems that hold up in production.

Building data pipelines in San Diego.


I.

Background

Grew up in a military family, moving between bases and overseas. Learned to adapt fast and play with the cards that were dealt.

Hobbies
  • Valorant
  • Poker
  • Hiking
  • Snowboarding
  • Surfing
  • Gym
  • Doomscrolling
  • Reading classics
  • Religious texts

II.

Work

  1. June 2026

    Researcher

    Calit2

    Built data workflows on CouchDB for a sensor and survey project.

  2. April 2026

    Data science researcher

    Smarr Lab, HDSI

    Rebuilt query paths over 700M+ rows in BigQuery, cutting runtime by 7x.

  3. 2023-2024

    Barista

    Starbucks at Target

    High school part time job.

  4. 2022

    Researcher

    JSHS and NASA

    Army DEVCOM student researcher. Presented findings at a national conference.

  5. 2021

    Mail clerk

    USPS, Camp Humphreys

    Ran high volume mail processing on a military installation abroad.


III.

Projects

Δt1 Δt2 Δt3 Δt4 s1 s2 s3 s4 32.84N 117.27W 41.1 kA 0 50 km N
fig. i · time of arrival across four stations

Plate I · 2025

Lightning strike data pipeline ↗

Reverse engineered Blitzortung's WebSocket binary protocol, then built ingestion with a custom decoder, PostGIS spatial indexing, FastAPI endpoints, and full Docker containerization.

  • WebSocket
  • Binary decoder
  • PostgreSQL
  • PostGIS
  • FastAPI
  • Docker
ipo interest time signals
fig. ii · interest over time, scored like an ipo

Plate II · 2026

Trends for Forum ↗

Deployed FastAPI app with SSE streaming, Google Trends ingestion, an HMAC authed market API, and an IPO style scoring engine. Built at the SDxUCSD Agent Hackathon.

  • FastAPI
  • SSE
  • HMAC
  • Scoring engine
  • Claude API
p99 latency slo degraded 1e4a7f0cb93d 2c6198ae40f2 d7c35a1be08f 3c9d culprit triage i ii iii brief
fig. iii · a buried bad commit, triaged in three steps

Plate III · 2026

Oncall agent ↗

Autonomous incident response agent. A chaos harness buries a bad commit in a live service's history and degrades it for real, then the agent correlates commits, diffs, log aggregates, and runbooks through three Claude triage steps, posts a structured brief to Slack, and writes the postmortem on resolve. 12/12 on the end to end eval, including a no culprit negative control.

  • Python
  • Claude API
  • Chaos injection
  • PostgreSQL
  • Slack
  • Docker

Reach out.

kevinkpyo@gmail.com