IEEE Agentic Formatter

What it does

One pipeline from unstructured text to publication-ready IEEE paper. No manual formatting. No template wrestling.

🏠

Local Mode — No API Key

Runs fully offline via Ollama (qwen2.5:7b, llama3, mistral). Zero cost, zero data sent to the cloud. Select Local (Ollama) in the sidebar.

🧠

Cloud Mode — Claude AI

claude-opus-4-8 with adaptive thinking and forced tool use for highest-quality extraction from complex or ambiguous inputs.

📄

Strict IEEE Layout

Letter-size, mirrored margins, single-column title block, continuous-break into two justified columns, Times New Roman typography, Roman-numeral headings.

📄

Multi-format Input

Paste raw text or upload a .txt, .md, or .docx file. The AI infers and fills missing sections from context.

IEEE Formatting Specification

Every output is validated against these constraints before download.

Property	Value
Page size	Letter (8.5 x 11 in)
Margins	1 in all sides, mirrored (odd/even pages)
Title font	Times New Roman 24 pt, centered
Author block	Times New Roman 10 pt (name) / 10 pt italic (affiliation)
Abstract / Keywords	Times New Roman 9 pt bold-italic label + 9 pt italic body
Body columns	2 equal columns, 0.32 in gap, continuous section break
Body font	Times New Roman 10 pt, justified, 0.15 in first-line indent, 12 pt exact leading
Section headings	Level 1: Roman numeral (I. INTRODUCTION) centered bold; Level 2: Letter (A. Background) left italic
References	Times New Roman 8 pt, hanging indent, IEEE bracket style

Local Mode (No API Key Required)

Run entirely on your machine — free, private, offline.

Install Ollama Download from ollama.com and start it.
Pull a model ollama pull qwen2.5:7b (recommended) or ollama pull llama3
Clone and install git clone https://github.com/sunilgentyala/ieee-agentic-formatter.git && pip install -r requirements.txt
Run — no .env needed streamlit run app.py — sidebar defaults to Local (Ollama)

ollama pull qwen2.5:7b
git clone https://github.com/sunilgentyala/ieee-agentic-formatter.git
cd ieee-agentic-formatter
pip install -r requirements.txt
streamlit run app.py
# Open http://localhost:8501 — Local (Ollama) is selected by default

Cloud Mode (Anthropic API)

Highest quality for complex inputs — requires an Anthropic API key.

Clone and install git clone https://github.com/sunilgentyala/ieee-agentic-formatter.git && pip install -r requirements.txt
Add your API key echo "ANTHROPIC_API_KEY=sk-ant-..." > .env
Run and switch backend streamlit run app.py — select Anthropic API in the sidebar
Paste or upload content and click Generate Download the IEEE-compliant .docx instantly.

Architecture

Three clean modules, zero circular dependencies.

agent/

models.py -- Pydantic v2 data models (IEEEPaper, Author, Section).

text_parser.py -- Two backends:
• parse_raw_text(): Anthropic streaming + forced tool_choice
• parse_raw_text_local(): Ollama OpenAI-compat endpoint + JSON-schema prompting

formatter/

docx_engine.py -- python-docx + raw OOXML (OxmlElement) to inject w:mirrorMargins, two-column w:cols, and exact-leading paragraph spacing without a Word template dependency.

Stack

Ollama (Local)

qwen2.5:7b, llama3, mistral via OpenAI-compatible endpoint. Fully offline, zero cost.

Claude claude-opus-4-8

Anthropic's most capable model with adaptive thinking and structured tool-use output.

python-docx + OOXML

Fine-grained Word document control via both the high-level API and raw XML injection.

Pydantic v2

Schema-validated structured output ensures the formatter always receives well-formed data.