IEEE Agentic Formatter

Transform raw research notes, drafts, or Markdown into a strictly compliant,
two-column IEEE Conference Paper (.docx) — in seconds.

Live Site |
Quick Start |
Local Mode |
Architecture |
IEEE Spec

What it does

One pipeline — from unstructured text to publication-ready IEEE paper:

Paste raw notes, a rough draft, or bullet points — or upload a .txt, .md, or .docx file
AI parses the content (local Ollama or Anthropic cloud) and extracts the full paper structure
python-docx + OOXML renders a pixel-perfect IEEE two-column layout
Download the .docx instantly from the browser

No API key required. Run fully offline with a local Ollama model.

Quick Start

# 1. Clone
git clone https://github.com/sunilgentyala/ieee-agentic-formatter.git
cd ieee-agentic-formatter

# 2. Install
pip install -r requirements.txt

# 3. Run  (no API key needed — uses local Ollama by default)
streamlit run app.py

Open http://localhost:8501, select Local (Ollama) in the sidebar, paste or upload your content, and click Generate IEEE Paper.

Local Mode (No API Key)

The app defaults to Local (Ollama) — zero cost, fully offline, no account needed.

Prerequisites

Install Ollama and pull a model:

ollama pull qwen2.5:7b    # recommended — best structured JSON output
# or
ollama pull llama3        # alternative

Sidebar options

Setting	Default	Notes
Backend	`Local (Ollama)`	Switch to `Anthropic API` for cloud
Local Model	`qwen2.5:7b`	Also supports `llama3`, `mistral`
Ollama URL	`http://localhost:11434`	Change if Ollama runs on another port

How it works

parse_raw_text_local() sends the raw content to Ollama's OpenAI-compatible endpoint (/v1/chat/completions) with a strict JSON-schema system prompt. The response is parsed, validated by Pydantic v2, and fed directly into the python-docx formatter.

Cloud Mode (Anthropic API)

Switch the sidebar to Anthropic API for claude-opus-4-8 with adaptive thinking and forced tool use — highest quality output for complex or ambiguous inputs.

echo "ANTHROPIC_API_KEY=sk-ant-..." > .env
streamlit run app.py

Or enter the key directly in the sidebar at runtime.

Architecture

ieee-agentic-formatter/
├── agent/
│   ├── models.py        # Pydantic v2: IEEEPaper, Author, Section
│   └── text_parser.py   # parse_raw_text() — Anthropic
│                        # parse_raw_text_local() — Ollama
├── formatter/
│   └── docx_engine.py   # python-docx + OOXML IEEE layout engine
├── docs/
│   ├── index.html       # GitHub Pages site
│   ├── favicon.svg      # Address-bar logo
│   └── og-image.svg     # Social preview banner
├── app.py               # Streamlit UI (backend selector)
└── requirements.txt

Module	Responsibility
`agent/models.py`	Pydantic v2 schema — `IEEEPaper`, `Author`, `Section`
`agent/text_parser.py`	Anthropic: streaming + `tool_choice` forced to `structure_ieee_paper`. Local: Ollama OpenAI-compat endpoint + JSON-mode prompting
`formatter/docx_engine.py`	python-docx + `OxmlElement` injection for `w:mirrorMargins`, two-column `w:cols`, exact-leading paragraphs
`app.py`	Streamlit: backend radio, model selector, text paste / file upload, live preview, download

IEEE Formatting Spec

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

Stack

Layer	Technology
Local LLM	Ollama (`qwen2.5:7b`, `llama3`, `mistral`) via OpenAI-compatible API
Cloud LLM	`claude-opus-4-8` with adaptive thinking + `tool_choice`
Structured output	Pydantic v2 + JSON-mode prompting / Anthropic tool use
Document generation	python-docx 1.1+ with raw OOXML injection
UI	Streamlit 1.40+
Config	python-dotenv

Releases

v1.1.0 — Local Ollama Backend (2026-06-26)

Local mode: run fully offline via Ollama — no API key required
parse_raw_text_local(): OpenAI-compatible Ollama endpoint + JSON-schema prompting
Streamlit sidebar: backend radio (Local / Anthropic), model selector, Ollama URL
Tested with qwen2.5:7b and llama3 on multi-section research drafts
openai>=1.0.0 added to requirements for Ollama client

v1.0.0 — Initial Release (2026-06-26)

Agentic parsing pipeline: Claude claude-opus-4-8 + forced tool_use + Pydantic v2 validation
Strict IEEE two-column .docx output via python-docx and OOXML
Multi-format input: plain text, .txt, .md, .docx upload
Streamlit UI with live structure preview and one-click download
GitHub Pages site with SVG favicon and social preview banner

Author

Sunil Gentyala

AI/ML Researcher | IEEE Member | HCLTech (HCL America Inc.)

Independent researcher specialising in agentic AI systems, multi-agent security frameworks,
federated learning, and quantum-safe protocols — with 8+ accepted IEEE publications.

Contributors

Sunil Gentyala
_{Creator & Maintainer}

ieee-agentic-formatter