Mulugeta Solomon AbateTOKYO, JAPAN

I build full‑stack systems for problems that matter.

End to end — the APIs, data models, and services (FastAPI, NestJS), the ML inside, and the cloud it runs on, with a React and Next.js front end on top. Flood‑simulation and disease‑risk platforms deployed across Africa, plus a SaaS side project for the East African market.

Currently · Software Engineer — Sora Technology, Tokyo

Email me GitHub LinkedIn

Based in Tokyo, Japan · open to on-site & remote

Works with

Selected work

Production systems in climate, public health, and hospitality — the problem, my role, the stack, and what shipped.

Climate · Disaster response

Flood-Sight

A physics-based flood-simulation and early-warning platform, deployed for disaster preparedness in Mozambique through a JICA-linked program.

My role

Built it end to end — the FastAPI + Celery backend, simulation-engine integration, geospatial data pipelines, and the Next.js + MapLibre front end — then made the solver 12.5× faster.

Deployed · MozambiqueSolver 12.5× faster146+ automated testsi18n · 4 languages

FastAPICeleryNext.jsMapLibre GLPostgreSQLRailwayflood-sight.com ↗

System architectureOne engineer · every layer

Rainfall and terrain in, a flood‑depth map out — end to end.

A Next.js front end talks to a FastAPI control plane that enqueues long‑running simulations onto Celery workers through Redis, streams live progress back over WebSocket, and tracks every job in Postgres. The flood engine — Landlab, SCS‑CN runoff, shallow‑water flow — writes GeoTIFF and PNG outputs to S3‑compatible storage (Cloudflare R2). JWT auth, transactional email, and Sentry observability round it out, deployed on Railway. A later profiling pass made the solver 12.5× faster per step, measured on staging, with byte‑identical output.

Click to view full size ↗

Frontend

Next.js 16 · React 19 · TypeScript · MapLibre GL · i18n (EN/FR/PT/JA)

Backend

FastAPI · SQLAlchemy 2 async · Pydantic · WebSocket progress · JWT auth

Data

PostgreSQL 16 · Alembic migrations · Redis 7 · async pipelines

Simulation

Python · Landlab · SCS‑CN runoff · shallow‑water (SWE) · rasterio

Infra

Railway · Docker · Celery · S3 / Cloudflare R2 · Resend · Sentry

Public health · Applied ML

LSM — Larval Source Management

AI web services that detect mosquito breeding sites from drone imagery and model disease risk — a greenfield services line supporting malaria prevention across multiple countries.

My role

Architecting the detection services and inference pipeline — models, evaluation frameworks, and AWS job orchestration.

>85% precision / recall+20% accuracymAP · IoU · F1 eval

PythonPyTorchYOLOv8SQSEC2 · GPUS3Internal platform

LSM inference job pipeline sequence diagram: client, FastAPI on ECS Fargate, RDS Postgres, S3, SQS, and a GPU worker on EC2

Request flow · click to enlarge ↗

Platform architectureSORA · AWS

The infrastructure the detection services run on.

A FastAPI control plane sits in a private subnet behind API Gateway; drone imagery lands in S3, and detection jobs are decoupled through SQS so GPU workers on EC2 pull and process them independently. Job state and results persist in RDS Postgres. The same async, queue‑backed pattern lets inference scale without ever blocking the API.

SORA platform AWS architecture: clients, API Gateway, Lambda, a FastAPI service in a private subnet, SQS, EC2 GPU inference, RDS Postgres, and S3

Click to view full size ↗

Side project · SaaS · Solo build

Karamu

A restaurant reservation and management platform for the East African market — designed, built, and shipped end-to-end: product, brand, and engineering.

My role

Side project, built solo outside work — product design, brand identity, full-stack build, and infrastructure.

Real-time floorM-Pesa paymentsSMS / WhatsApp

ReactNext.jsTypeScriptNodeNestJSPrismaSocket.iogetkaramu.com ↗

System architectureSolo build · every layer

A reservation system that has to be right on a Friday night.

One system, five surfaces: reservations, a restaurant dashboard, table management, QR menus, and a kitchen display system (KDS). A Next.js front end drives them against a NestJS API, with everything modeled in PostgreSQL through Prisma. Socket.io keeps the floor and the kitchen in sync in real time, M‑Pesa handles payments, and confirmations go out over SMS and WhatsApp.

Frontend

React · TypeScript · Next.js · Tailwind · real‑time floor UI

Backend

NestJS · REST · Better Auth · webhook handlers

Data

PostgreSQL · Prisma ORM · migrations

Real‑time

Socket.io · live floor + hold state

Integrations

M‑Pesa payments · SMS / WhatsApp

Infra

Cloudflare Workers · R2 · CDN · Railway · Docker · CI/CD · getkaramu.com

About

Robotics gave me systems thinking. Now I point it at the whole stack.

I started in robotics — an M.Sc. at Ritsumeikan, a teaching assistantship, and a published paper on semantic line detection. Control theory, math, and sensor systems are where my instinct for infrastructure comes from.

Today I work across the whole stack: the interface people touch, the APIs and data models behind it, the ML that powers the hard parts, and the cloud that holds it up. I’ve written production SQL against real datasets, shipped real‑time features, and owned services end to end — from schema design to deploy. Startup pace taught me to go deep fast: not a little of everything, but whole systems, built to hold.

I grew up in Ethiopia and build with the East African market in mind — software for the places default tools forget. I care about systems that hold up when the stakes are real: a flood, an outbreak, a restaurant's Friday night.

Journey

From a bachelor's in Ethiopia to graduate robotics in Japan to shipping production ML — the path in five steps.

2016 – 2021

B.Sc. Electromechanical Engineering

Addis Ababa Science & Technology University · Ethiopia

Graduated 3.81 / 4.0. Foundations in embedded systems, control, data structures, and computer vision.

2022 – 2024

M.Sc. Robotics

Ritsumeikan University · Kyoto, Japan

MEXT Scholar. Robot control, AI, and security. Teaching assistant for 60+ graduate students, and co-authored an RSJ2024 paper on semantic line detection.

MEXT ScholarRSJ2024

2024 – 2025

A2SV — Backed by GoogleConcurrent

Competitive Programming & Software Engineering

1,500+ hours of algorithms and industry engineering practice with a program that has placed 80+ engineers at Google, Amazon, and Bloomberg.

Jul – Oct 2024

Software / ML Intern

Sora Technology · Tokyo, Japan

Optimized drone path-planning (−33% acquisition time) and migrated object detection from YOLOv5 to YOLOv8 (+22% accuracy).

Oct 2024 – Present

Software EngineerNOW

Sora Technology · Tokyo, Japan

Architected the SORA platform backend on AWS, built Flood-Sight end-to-end and made its flood solver 12.5× faster, and shipped waterbody AI detection at >85% precision / recall. Now lead developer on a disease early-warning platform for Ghana.

AWSFlood-SightLSMEarly warning

Stack

Languages

Python · TypeScript · SQL — working knowledge of Go, C++, C

Frontend

React · Next.js · Vite · Tailwind · MapLibre GL

Backend

Node / NestJS · FastAPI · Celery · REST · WebSockets · JWT auth

Data / ML

PyTorch · TensorFlow · YOLO · OpenCV · scikit-learn · Data pipelines · Geospatial

Cloud / Infra

AWS — Lambda, API Gateway, SQS, RDS, S3, EC2 · Railway · Docker · CI/CD · PostgreSQL · Redis · Sentry

05Contact

Let's build something that matters.

Open to full‑stack and backend engineering roles where the work has real stakes — and the occasional ambitious collaboration.

Based in Tokyo, Japan · authorized to work in Japan · open to remote

[email protected]

GitHub LinkedIn Resume