“Heterogeneous Agentic AI for UI Code Generation”
A research project exploring fine-tuned Small Language Models (SLMs) in a multi-agent architecture for production-ready React/Next.js landing page generation.

Overview

This repository contains two main components:

• /academic — paper, LaTeX source, dataset documentation, and finetuning notebooks
• /app — Production implementation: Electron desktop app, FastAPI server for Langraph agent, and Next.js preview toolkit

Key Innovation

Instead of using a single large model for everything, Qwendean uses specialized models for specialized tasks:

• Gemma 4 E2B (2.3B) — handles conversation, requirements gathering, and task decomposition
• Qwendean (fine-tuned Qwen3-4B) — generates ShadCN/Tailwind/React components with high consistency

Result: 3-5× faster generation, lower cost, better output consistency than general-purpose LLMs.

Quick Start

Prerequisites

• Node.js 20+, Python 3.11+, Ollama
• Models: hf.co/iamdyeus/qwendean-4b-GGUF:latest and gemma4:e2b

Paper Abstract

This work introduces a heterogeneous agentic AI architecture using fine-tuned SLMs for efficient UI component generation. Unlike monolithic LLM approaches, we separate orchestration (lightweight general-purpose model) from execution (domain-specialist model). Fine-tuning used LoRA with rsLoRA at 16-bit precision on 4,100+ samples. Empirical results show better cohesion, consistency, and 3-5× latency reduction compared to homogeneous baselines.

Citation

If you use this work, please cite:

@misc{qwendean2026,
  title={Heterogeneous Agentic AI System Using Fine-Tuned SLMs},
  author={Arsh, Swarnendu Ghosh},
  year={2026},
  note={Available at \url{https://github.com/iamDyeus/qwendean}}
}