Naol Tamrat
Building AI agents and production web apps, from LangGraph pipelines to the Next.js products people use every day
About Me
Full-Stack Developer and AI Engineer with experience building production systems across the stack — from Next.js frontends to Django APIs to LLM-powered pipelines with LangChain and LangGraph.
I'm an A2SV fellow (top 1% of applicants, Google-backed), solved 600+ problems on LeetCode, and hold a BSc in Software Engineering with a 3.63 GPA and Presidential Award.
Currently building medical document processing APIs and AI pipelines at Hydrus.ai. Previously cut LLM costs by 67% at Fluid AI and led frontend development at Eskalate.
Languages
Frontend
Backend
AI / ML
Tools
Experience
Where I've worked and what I've built
Vella AI
Contract Full-Stack AI Engineer
May 2026 – Jun 2026 · Remote
Built most of Arena, an AI campaign intelligence platform for PPC teams, end to end: Campaign Coach, Budget Simulator, Seasonality Planner, an ML budget allocator, and analytics and benchmark dashboards.
Hydrus.ai
Full-Stack & AI Engineer
Oct 2025 – May 2026 · Remote
Attian: Migrated a Framer marketing site to Next.js with pixel-perfect design and motion across 8+ page templates; built a 5-schema Sanity CMS that lets the team publish articles, case studies, and reports with embedded charts and graphs, plus automated lead capture and ISR.
Fluid AI
AI Engineer & Backend Developer
Feb 2025 – Aug 2025 · Dubai (Remote)
Cut LLM API costs by 67% and response time from ~9s to ~3s by building a response interception system that bypasses LLM calls for known structured outputs from tools like Maps, Calendar, and Hotels.
Eskalate
Project Lead & Frontend Developer
Aug 2024 – Jan 2025 · Remote
Led product direction for eskalate.io, planning roadmaps and running Agile sprints based on 90+ developer interviews.
Eskalate
Frontend Developer
Oct 2023 – Jul 2024 · Remote
Architected the frontend from scratch: folder structure, Redux/RTK Query state layer, refresh-token system, coding guidelines, linters, CI/CD, and GitHub configuration.
Freelance
Web Developer
Jan 2024 – Apr 2024 · Addis Ababa, Ethiopia
Designed and built three responsive websites, two for ministry organizations and one for a coffee export company, each tailored to its audience.
Projects
Things I've built and shipped
Multi-tenant inventory and sales SaaS for small retail shops in Ethiopia, live and used by 3 shops since July 2026, with a trilingual AI assistant.
- • Products with variants, sales and returns, stock receiving and corrections, reports, staff accounts, public product catalogs, and paid Basic/Pro plans.
- • "Ask IMS": a LangGraph + Gemini assistant that answers shop owners' questions in English, Amharic, or Afaan Oromo using 8 read-only data tools, with every number traced to a tool result and an evaluation suite to check accuracy.
- • Server-side tenant isolation and role-based access (owner/staff/admin); transactional stock changes with row locking to prevent overselling, plus a full audit log.
- • 258 integration tests and 79 Playwright end-to-end tests in CI, with automated migrations, 6-hourly backups, and Sentry monitoring.
Expense-sharing and personal finance app used by 10 people to log 680+ expenses since April 2026, recently rebuilt as a mobile-first PWA.
- • Group and person-to-person expenses, loans, balances and settle-up, recurring bills, and monthly spending summaries.
- • TypeScript monorepo (Turborepo) with a tested money domain using integer minor units, a validated REST API, PostgreSQL with Drizzle and transactional writes, and Better Auth with email verification and rate limiting.
- • Covered by 399 automated tests plus Playwright end-to-end checks on phone viewports in CI.
LangGraph agent system with MCP tool integrations, built at Fluid AI and open-sourced with permission.
- • Modular agents with tools served as MCP-based microservices and LangGraph-managed state.
- • Swarm-style architecture with dynamic task handoff between agents.
FastAPI service running parallel browser agents for real-world restaurant reservations, built at Fluid AI and open-sourced with permission.
- • Automated reservations across Google Maps, OpenTable, and SevenRooms with browser-use.
- • Parallel Chrome instances with persistent profile management.
AI-powered automation platform that processes forwarded emails, extracts unsubscribe links, and uses an LLM-controlled browser agent to unsubscribe automatically.
- • Dual extraction strategy: HTML parsing first, LLM fallback for complex emails, reducing API calls by 90%.
- • Domain learning system that caches successful unsubscribe URLs for instant future processing.
- • Celery + Redis task queue with a FastAPI webhook for async email processing.
Let's Work Together
Open to full-stack, backend, and AI engineering roles. Let's build something great.