目录

Hi there 👋

I’m Arman Radmanesh

Graduate Research Assistant | MS Data Science & Analytics @ University of Oklahoma

Master’s student in Data Science and Analytics with 10+ years of software engineering experience, specializing in Agentic AI workflows, Prompt Engineering, RL-based LLM post-training, Retrieval Augmented Generation (RAGs), Coding Agents, and multi-agent systems.

📫 Contact: radmanesh@ou.edu | LinkedIn | Website


🔬 Current Research

Graduate Research Assistant at University of Oklahoma (Dec 2024 – Present)

  • Researching multi-agent code generation and evaluation with emphasis on agentic AI workflows and A2A orchestration
  • Developing pipelines for RL-based LLM post-training, semantic retrieval, and domain-specific RAG systems
  • Building robust multi-agent environments for iterative code generation, tool calling, and structured benchmarking
  • Working on prompt engineering strategies, multimodal layout evaluation, and reproducible system-level testing frameworks

🎓 Education

  • Master of Science in Data Science and Analytics - University of Oklahoma, Norman, OK (Graduating Dec 2026)
  • Bachelor of Computer Engineering (minor in IT) - Amirkabir University of Technology, Tehran, Iran (2011)

💻 Skills

Programming & Frameworks

Python, Java, Spring, C++, JavaScript/TypeScript, ReactJS, Next.js, Node.js, Express.js, HTML5, CSS3, Sass, jQuery, Tailwind CSS, Webpack, Grunt, Bun

AI & Machine Learning

LLM inference & fine-tuning, RL-based post-training, RAGs, Prompt Engineering, Agentic AI workflows, multi-agent pipelines, testing/benchmarking, data preprocessing

Other Technologies

SQL and NoSQL Databases, Unix command-line tools, Git, Docker, API design and RESTful services

Languages

English, Farsi


🚀 Notable Projects

AgentifiedDesign2Code

A multi-agent extension of the Design2Code pipeline created as part of the Berkeley AgentBeats Competition. This system enables a generation agent to produce HTML and CSS while an evaluation agent scores layout fidelity using Playwright, geometric IoU analysis, and multi-device screenshot comparison, and the fully agentified A2A workflow reproduces original Design2Code benchmarks while supporting structured critique-and-repair loops and tool-driven reasoning workflows.

ResponsiveGen: Multi-Device Sketch-to-Code System

A vision-enabled LLM pipeline that converts mobile, tablet, and desktop wireframe sketches into fully responsive HTML and CSS, featuring an evaluation suite that integrates screenshot comparison, geometric layout analysis, and LLM-as-judge rubric scoring to benchmark how effectively multimodal models generate production-quality responsive layouts.

Intelligent Faculty Recommendation Engine (OU DSAI)

An LLM-powered semantic search and ranking system that matches PhD proposals to the most relevant faculty across OU, using LlamaIndex, LangChain, LangGraph, Qwen models, and ChromaDB to ingest CVs, publications, and proposal documents and perform expertise extraction, vector retrieval, and accurate ranking at scale.

Jamasp

A scalable platform for wearable-device research with participant management and AI-enabled analytics. Built for behavioral science research with real-time data collection and analysis capabilities.

Lens (Cut)

A mobile-friendly toolkit (ReactJS, MongoDB) for behavioral experiments including ultimatum games, memory tests, and fasting studies. Designed for social research applications.


💼 Professional Experience Highlights

  • Graduate Research Assistant – University of Oklahoma (Dec 2024 – Present)
    Researching multi-agent code generation and evaluation systems, RL-based LLM post-training workflows, semantic retrieval pipelines, and multimodal sketch-to-code models. Building agentic AI tools and evaluation frameworks using Python, PyTorch, LangChain, LangGraph, and vision-enabled LLMs.

  • Software Developer and Researcher – New School for Social Research (Aug 2022 – Feb 2025)
    Developed Jamasp and Lens, large-scale research platforms for behavioral science experiments and wearable analytics.

  • Senior Software Engineer – BongaMonga (2016 – 2018)
    Developed and maintained a real estate web application using modern full-stack technologies.

  • Senior Software Engineer – WhiteOx (2015 – 2016)
    Built multiplayer card game UI, poker analysis tooling, and a reward system that significantly increased user engagement.

  • Senior Software Engineer / Co-founder – Ratnic (2014 – 2015)
    Designed and developed a quiz-based social platform and led engineering for an optimized flight ticket search engine.

  • Research Engineer and Developer – ICT Research Institute (2012 – 2013)
    Worked on ontology-based knowledge engineering, semantic processing, and early AI-driven research tools.


🌐 Cloud & Infrastructure

I frequently use Google Cloud Platform services:

  • Firebase (Hosting, Authentication, Functions, Firestore, Extensions)
  • Cloud Logging, Cloud Run, BigQuery, Vertex AI

Additional platforms and tooling I use:

  • Lambda Labs (Lambda AI) for GPU compute, model training, and large-scale experimentation
  • Weights and Biases (WandB) for experiment tracking, model logging, and reproducible ML workflows
  • Phoenix (Arize AI) for LLM evaluation, model observability, and dataset quality monitoring
  • OSCER HPC (OU Supercomputing Center) using SLURM for distributed training, GPU jobs, and large-scale ML pipelines
  • Delta AI for high-performance compute tasks, GPU acceleration, and scalable model experiments

📖 Interests

I love to read, both fiction and non-fiction, and I’m always exploring new technologies and research methodologies in AI and software engineering.


⚡️ Recent Activity

Last Updated: Sunday, March 22nd, 2026, 12:11:34 PM


关于
5.3 MB
邀请码