Applied AI Engineer

Mahdi Abbasi

I build agentic AI systems, RAG pipelines and NLP products, and ship them as real, deployed services.

Rasht, Iran · Open to remote / hybrid

About

Applied AI engineer with 6+ years of hands-on experience in Natural Language Processing, agentic AI, Retrieval-Augmented Generation, data engineering and Docker-based deployments. I focus on turning AI technologies into practical products that solve real business problems: designing intelligent applications, automating workflows and shipping scalable AI-powered services. Active competitor in national AI competitions and hackathons.

6+
years in AI
2026
Silver, AIROCUP
6
hackathon & challenge builds

Experience

Artificial Intelligence Engineer · Irsa (part-time)

  • Design and build AI-powered applications using NLP, LLMs and agent-based architectures.
  • Develop RAG pipelines that connect language models to company data for search and question answering.
  • Build data processing pipelines to clean, transform and prepare data for ML models and AI services.
  • Automate business workflows with AI agents and deploy services as Docker containers for reliable, scalable delivery.

Honors & competitions

Silver Medal · Intelligent Agent League, AIROCUP 2026

National AI & Robotics Competition. Designed and built an autonomous AI agent for a competitive environment, focused on agent design, decision-making and real-time problem solving.

Agentic AI Hackathon · Starcoach

72-hour hackathon building agent-based solutions to real-world problems set by Torob, Liara and Wallex. Also took part in the Starcoach Vibe Coding Hackathon (Sep 2026).

GCI World Program · Matsuo-Iwasawa Lab, University of Tokyo

Accepted into the international AI and data science program.

3rd Place · Suspicious Banking Transaction Detection, AAIC

7th Amirkabir AI Competition (Amirkabir University of Technology). Built an ML model to detect fraudulent and suspicious transactions in banking data.

Projects

Built for hackathons and real challenges. Everything is open source, with a demo video for each.

01

Botyar

Describe your bot in Persian, get a working one

A business owner describes a Bale or Telegram bot in plain Persian. A LangGraph builder agent asks about ambiguities, designs and builds the bot, writes and runs its own tests, publishes the tested version, and applies later changes with a diff and a full test re-run. Built for the BuildX event.

  • LangGraph
  • FastAPI
  • PostgreSQL
  • Next.js
  • Bale / Telegram

02

Liara Launchpad

An agent that deploys your finished project

Reads a project, decides what Liara needs (apps, databases, disks, env vars), shows a plan with evidence and price, and deploys only after approval. It verifies the app responds, and when something fails it classifies the cause and offers a fix. Fixes that cost money are never applied without asking.

  • LLM agent
  • Tool use
  • Liara API
  • Docker

03

Torob In-Person Shopping Assistant

Search over 226k products in ~10 ms

A mobile, RTL shopping assistant built for the Starcoach hackathon on Torob data. FastAPI backend over a single SQLite file with FTS5 Persian full-text search, plus a Next.js frontend.

  • FastAPI
  • SQLite FTS5
  • Next.js
  • Persian search

04

CodeBiome

Verified architecture maps from any GitHub repo

Paste a repository URL and an AI agent you connect over MCP (Claude, Cursor, Cline and more) explores the code and submits modules, flows and health findings. Every file path and module reference is checked against the real file tree; invented claims are rejected.

  • MCP
  • TypeScript
  • Agents
  • Next.js

05

Liara Docs Assistant

Chat that opens the exact docs section it cites

RAG assistant for Liara documentation with hybrid retrieval (lexical + semantic embeddings, fused with Reciprocal Rank Fusion). Cited pages open in a side panel that scrolls to and highlights the passage. Streaming answers via AI SDK v5.

  • RAG
  • Hybrid search
  • AI SDK
  • TypeScript

06

Zarinpal Merchant Analytics

Agent-based, actionable insights per merchant

Built for the Zarinpal data challenge on 2.2M payment attempts across 343 merchants. Funnel by payment provider, retry economics and peer-percentile comparisons, with every number tracing back to where it came from.

  • Agents
  • Analytics
  • TypeScript
  • 2.2M rows

More on GitHub

Skills

AI / LLM
Agentic AI development, RAG, LangChain, LlamaIndex, Anthropic Claude, Claude Code & subagents, Claude Skills, MCP
ML & NLP
Machine learning, Natural Language Processing, data processing, Pandas
Engineering
Python, Docker
Languages
Persian (native), English (professional working proficiency)

Education

B.Sc. Computer Engineering
Payame Noor University
Sep 2019 – Dec 2026 (expected)

Certifications

  • RAG for Production with LangChain & LlamaIndex · Activeloop, 2026
  • Claude Code 101 · Claude 101 · Anthropic, 2026
  • Intermediate ML · Intro to ML · Pandas · Python · Kaggle, 2023
  • Programming with Python · Tehran Institute of Technology, 2021

Let’s talk

Open to remote and hybrid roles building applied AI. The quickest way to reach me is email.

mahdi.abbasi.5254@gmail.com