$ measure01/04
Measure first
Traces and real numbers before opinions. The bottleneck is rarely where we guess.
$ whoami
Kai Nakamura
Backend Engineer_
I build the quiet systems that keep products fast at 3 a.m. — queues, APIs and infrastructure that scale without drama.
Tokyo, Japan●Open to new roles
hero//identity.jpg● online
- 7 yrs
- Experience
- 2.4B
- Requests / day
- 9.8k
- OSS stars
README.md
The short version.
Who I am and what I work on, before the details.
kai//portrait.jpg● loaded
Backend engineer with seven years of building high-throughput services in Go and Rust. I lead platform work at a cloud start-up, where I've cut p99 latency by 60% and moved a monolith to event-driven services without a single customer-facing outage. I care about observability, boring deploys and documentation that people actually read.
- Based in
- Tokyo, Japan
- Focus
- Backend & Infrastructure Engineer
- Currently
- Kumo Cloud
- Stack
- Go · Rust · TypeScript
./build.sh
How the work gets done.
Four stages, start to finish. Scroll through the pipeline.
git log --author=kai
Career, as a commit history.
Each role logged as a commit, with the work that went into it.
- #945cca0branch: mainApr 2023 — Present
Senior Backend Engineer
Kumo Cloud
- ›Lead a four-person platform team running 140 Go services on Kubernetes
- ›Redesigned the job queue, cutting p99 latency from 900ms to 350ms
- ›Introduced OpenTelemetry tracing and SLO dashboards across the company
- #ee47c23branch: relay-payJun 2019 — Mar 2023
Software Engineer
Relay Pay
- ›Built idempotent payment APIs handling 30M transactions a month
- ›Migrated a PostgreSQL monolith to event-driven services with Kafka
- ›Mentored six engineers through their first on-call rotations
/projects
Two builds worth a closer look.
Selected work: what it does, how it was built, and the stack behind it.
Queuebird
An open-source distributed job scheduler with exactly-once delivery and a tiny operational footprint.
- ›Handles 50k jobs per second on three nodes
- ›Used in production by 200+ companies
GoRedisKubernetesgRPC
02
Tracewise
Finds the most expensive spans in OpenTelemetry data and suggests where to sample.
- ›Cut tracing costs by 45% for early adopters
- ›Runs as a single static binary
RustClickHouseOpenTelemetry
~/research
Research, on the record.
Two publications.
60%
lower p99 latency
Tail-Latency Budgets for Multi-Tenant Job Queues
How per-tenant latency budgets and adaptive backpressure keep noisy neighbours from hurting everyone else.
- Distributed systems
- Performance
- SRE
18%
lower cloud spend
Forecasting Capacity With Seasonal Traffic Models
A practical guide to predicting infrastructure demand so autoscaling starts before the traffic arrives.
- Capacity planning
- Forecasting
package.json
The dependency manifest.
Tools and technologies, grouped by where they fit.
Languages
GoRustTypeScriptSQL
Infrastructure
KubernetesTerraformAWSGCP
Data
PostgreSQLRedisKafkaClickHouse
Practices
SREObservabilityAPI designIncident response
origin/education
Where the foundation was built.
Academic background and qualifications.
Academic background
- B.Eng. in Computer Science2019Keio University, Tokyo
Research
cd ./research →2 publications
Keio University — Class of 2019
./gallery
A few frames off the record.
Click any frame to open it full-size.
$ play --videos
Press play on the work.
Two reels: talks, demos and walkthroughs. Nothing loads until you hit play.
~/reels/01-featured-video.mp4featured
Featured video
Example video from YouTube. Add your own talks, product demos or reels here.
- ~/reels/02-another-example.mp4
Another example
Every theme shows videos with a lightweight click-to-play player.
$ contact --init
Building something? Let’s talk it through.
Open to new opportunities. Reach out through any of the channels below.
$ connect --profiles