About
I'm based in Addis Ababa, Ethiopia, working at the intersection of software engineering and AI. I like moving between disciplines: some days writing Python pipelines, other days building a UI, other days reviewing a dataset entry for the fifth time because something feels off. I like that variety, and I'm good at all three.
Right now I spend most of my time building and reviewing coding datasets that go directly into LLM training pipelines. Before that, I was in India, building and deploying production systems in the fintech space.
On the dev side, I build full-stack applications with React, Next.js, Vue, TypeScript, Node.js, and Express, backed by PostgreSQL or MongoDB and styled with Tailwind CSS. I pick the stack based on what the problem actually needs. I've built systems with a Next.js frontend and a dedicated Python backend, splitting the UI layer from the data/ML-heavy logic where Python made more sense. Be it a web app, a training pipeline, or the dataset that makes a model actually work, I care about the small details that separate something that runs from something that's genuinely good.
Experience
2025 - Present Coding Dataset Engineer & Project Lead · Eagle Point AI
Addis Ababa, Ethiopia
I build the data that teaches models how to code. Every task across Python, JavaScript, SQL, algorithms, and debugging gets written, tested, and reviewed like production code, because a bad example doesn't just fail once, it fails every time the model learns from it. Beyond my own datasets, I manage delivery for external partners, translating requirements from clients like ByteDance's Seed model team and Mindflow into specs engineers can execute against, then reviewing the output myself before it ships. That means catching the subtle stuff across other people's work too: an off-by-one in a spec, a constraint quietly dropped in a rewrite, an edge case nobody tested.
- Python
- JavaScript
- SQL
- Prompt Engineering
- LLM Fine-tuning
- Quality Review
2025 - Present Freelance Full-Stack Developer · Upwork
Remote
Long before "full-stack" was a title, it was just how I worked: clients show up with a rough idea, sometimes barely a sketch, and I take it from there. Over the years that's meant dashboards, internal tools, client-facing apps, API integrations, and more than a few projects where the real job was untangling someone else's half-finished codebase before building anything new. React or Vue on the front end, Node.js and Express underneath, PostgreSQL or MongoDB holding the data. The stack changes with the problem, but the standard doesn't: I ship things that still work six months later, not demos that impress once and break in production.
- React
- Vue
- TypeScript
- Node.js
- Express
- Python
- PostgreSQL
- MongoDB
2023 - 2024 Model Training Team Lead · Haidata
Karnataka, India
Led quality review across medical, coding, and multimodal datasets, the kind of work where a generic annotator sees an answer that looks right, and a domain expert sees the mistake hiding inside it. I built the standards the team worked against and trained new hires to see what I saw: the difference between an answer that looks correct and one that actually is.
- Quality Review
- Data Annotation
- Label Studio
- Multimodal
- Medical Datasets
Projects
Internal Platform · Eagle Point AI
Aquila
AI data production platform where I build, review, and validate coding datasets for LLM training, from task creation through quality review inside a secure, no-copy-paste contributor environment.
Internal tool (case study only)
- Python
- JavaScript
- LLM Datasets
- Quality Review
- Prompt Engineering
Featured Project
Future Quest
Built solo for a founder with a clear problem: thousands of students trying to navigate scholarships, work permits, and relocation across 30+ countries with no single place to make sense of it. I built the full product front to back: opportunity discovery, 1-on-1 consultation booking synced to live calendar slots, a study materials storefront, and a free open-resources hub. Trust was the hardest part to design for, not the calendar sync. When someone's deciding whether to hand a platform their next country, every section has to earn it. It's now helping 2,000+ students do exactly that.
- Next.js
- TypeScript
- Tailwind CSS
- Vercel
- Polar
Featured Project
PINNACLE AI
A creative agency running development, design, and consulting under one roof came to me needing a site that didn't just describe what they do. It had to prove it. I built it on Vue 3 with static-site generation so every service page loads instantly, added an animated client logo marquee and scroll-triggered impact counters to make their numbers land, and shipped a Formspree-backed contact pipeline with an accessible FAQ accordion. Deployed on Vercel, live, converting.
- Vue 3
- TypeScript
- Tailwind CSS
- Vite SSG
- Vercel
Featured Project
Frame by Frame
A video and photo production studio based in Addis Ababa, working with clients worldwide, needed a landing page as cinematic as their actual work. Anything less would've undersold them. I built the hero motion with GSAP ScrollTrigger and added a lazy-loaded Three.js 3D camera interaction for depth without the load-time cost. On the backend, the contact form ships with honeypot protection, rate-limiting, and production security headers, because a site for people who make things professionally can't afford to look or behave like a template. All on vanilla HTML/CSS and Vite: no framework weight where none was needed.
- HTML
- CSS
- Vite
- Three.js
- GSAP
Skills
Technologies I work with across full-stack development and AI data production.
Full-Stack
- React
- Next.js
- Vue
- TypeScript
- Python
- Node.js
- Express
- PostgreSQL
- MongoDB
- Tailwind CSS
AI & Data
- LLM Fine-tuning
- Prompt Engineering
- Dataset Curation
- SQL
- FastAPI
- Quality Review
- Label Studio
Tools & Cloud
- Git
- Docker
- AWS
- Vercel
- Linux
AI-assisted development
I use AI the same way I use any other tool: with intent, context, and a standard for the result. Claude, Codex, Cursor, and GitHub Copilot help me move faster through research, implementation, and review—not hand over the thinking. I validate the output, understand the trade-offs, and stay responsible for what ships.
- Claude
- Codex
- Cursor
- GitHub Copilot
Contact
Open to freelance work, AI roles, and anything genuinely interesting.