Anatoli Abahumna

Software engineer — systems, end to end

Anatoli Abahumna

I build practical software across web, iOS, cloud, operations tooling, and computer vision.

View work Research
CLIENT Web app CLIENT iOS app CAPTURE Telegram SERVICE HTTP API LOCAL SQLite CLOUD Supabase SERVICE Vision INTERNAL Ops desk FIG. 00 — THE SHAPE MOST OF THIS WORK TAKES CLIENTS → API → STORES CLIENT Web app CLIENT iOS app CAPTURE Telegram SERVICE HTTP API LOCAL SQLite CLOUD Supabase SERVICE Vision INTERNAL Ops desk FIG. 00 — THE SHAPE MOST OF THIS WORK TAKES

WebiOSCloudOps toolingComputer vision

YeneDesk

A personal productivity workspace built across web, API, iOS, local data, cloud beta, and operations tooling.

The work spans a React/Vite web app, Node/Express API, SQLite local mode, Supabase Auth and Postgres RLS cloud mode, SwiftUI iOS, BYOK AI integrations, Telegram capture, and an internal ops dashboard.

Product surface
Tasks, notes, finance, meals, Telegram capture, and AI-assisted review.
Platform work
Local-first desktop web, cloud beta, API layer, and SwiftUI iOS companion.
  • React / Vite
  • Express API
  • SQLite
  • Supabase RLS
  • SwiftUI iOS
  • Ops dashboard
M-01Tasks
M-02Notes
M-03Finance
M-04Meals
M-05Capture
M-06Review
Web iOS Telegram API SQLite Supabase
FIG. 01 — One workspace, six modules, one API underneath

Product work with working edges.

Each project is framed around what was actually built: product surface, data model, deployment path, and verification evidence. The figures are deliberately abstract — diagrams of how each system thinks, not screenshots.

P-02 · SwiftUI product system

YeneFitness

Native SwiftUI fitness app with SwiftData persistence, AI coach, workout import, nutrition extraction from meal photos, progress tracking, supplements, backup/restore, and XCTest/XCUITest coverage.

Mobile app
Workout, progress, supplement, and backup flows.
Meal photo work
Nutrition extraction lives inside the fitness product rather than a separate product.
  • SwiftUI
  • SwiftData
  • OpenAI
  • XCTest
TRAIN Sessions logged FUEL Meals from photos TREND Progress over time EFFORT / WEEK FIG. 02 — TRAINING MODEL, ABSTRACTED

P-03 · Public site + internal tooling

Anatoli Seeds + Operations Desk

Procurement-focused public website and internal RFQ, proforma, order, dispatch, reservation, quality-record, supply-planning, and operations-log tooling for seed supply workflows.

Commercial workflow
RFQ intake, proforma handling, order state, and dispatch review.
Operations surface
Reservations, quality records, supply planning, and activity logs.
  • Astro
  • Vite
  • Express
  • SQLite
STAGE 1 RFQ STAGE 2 Proforma STAGE 3 Order STAGE 4 Dispatch RUNNING UNDERNEATH Reservations HELD / RELEASED Quality records LOT / GERMINATION Supply planning SEASON / STOCK FIG. 03 — ONE ORDER, END TO END

Utilities & smaller systems

Focused tools with one job each, built to be used daily.

IN · MIC OUT · HEADPHONES ROUTE Output switch FIG. 04 — “SWITCH TO HEADPHONES”, SPOKEN

P-04 · macOS utility

YeneSound

Sound input/output control utility with voice-command shortcuts: switch devices, set levels, and mute without opening system settings.

  • Swift
  • Core Audio
  • Speech
13 MONTHS 12 × 30 + PAGUME ጳጉሜ FIG. 05 — A YEAR WITH A THIRTEENTH COLUMN

P-05 · Calendar system

YeneCal

Ethiopian calendar system: Ethiopian–Gregorian date conversion, the 13-month year with Pagume, holidays, and a widget-friendly month view.

  • SwiftUI
  • WidgetKit
  • Date math
BOARD STATE MOVE 14 · SOW → FIG. 06 — RULES ENGINE + MINIMAX OPPONENT

P-06 · Game system

Gebeta

Digital take on Ethiopian mancala: a rules engine with legal move validation, two-player mode, and a minimax AI opponent.

  • TypeScript
  • Canvas
  • Minimax
SOURCE → OPTIMIZED WEBP / AVIF HERO.PNG GALLERY.PNG COVER.JPG PLATE.PNG FIG. 07 — SAME PICTURE, LESS OF IT

P-07 · CLI tool

Presswork

Batch image-optimization CLI: converts sources to WebP/AVIF, resizes for target layouts, and reports size deltas before anything gets committed.

  • Node
  • Sharp
  • CLI

Smart Medicine Box Vision System

Blister-pack pill counting with an explainable OpenCV baseline, a YOLO11 detector path, a local FastAPI service, dataset tooling, and visual debug artifacts.

  • FastAPI
  • OpenCV
  • YOLO11
  • pytest
  1. 01
    Image normalization

    Prepare one blister-pack image for consistent downstream checks.

  2. 02
    Cavity detection

    Locate candidate regions with classical computer vision and model-based paths.

  3. 03
    Classification

    Return present, missing, low-confidence regions, timings, and warnings.

  4. 04
    Debug outputs

    Produce overlays and galleries that make failures inspectable.

PLATE SCAN 12 CAVITIES MISSING CONF 0.42 RESPONSE PRESENT / MISSING / LOW-CONF + TIMINGS FIG. 08 — FAILURES MADE INSPECTABLE