## What it does

Inkling turns a single topic into a fully researched, fact-checked report — no manual searching, reading, or drafting required. Give it a topic, and four specialized AI agents take it from there: one searches, one reads, one writes, and one critiques.

## How it works

Inkling isn't one model doing everything — it's four agents handing work off to each other, each with a single job:

- Search Agent — finds recent, reliable sources on the topic

- Reader Agent — picks the most relevant source and scrapes it for depth

- Writer Chain — drafts a structured report from the combined research

- Critic Chain — reviews the report and flags gaps or weak claims

The full shared state (search results, scraped content, report, and feedback) is returned at the end, so nothing along the way is thrown away.

## Tech stack

- LLM reasoning — Mistral AI via langchain-mistralai

- Agent orchestration — LangGraph / LangChain

- Web search — Tavily

- Content extraction — requests + BeautifulSoup

- UI — Streamlit

## Architecture

Each stage is invoked, writes its output into one shared state dict, and hands off to the next — nothing is re-derived, and the UI just reads the final state.

## Roadmap

- Stream live per-agent progress instead of a single blocking run

- Support multiple scraped sources per report, not just one

- One-click export to PDF/Word

- Swap in alternate LLM providers
Special Thanks to AyushGautam(Cse-Ds)and Kaushik kumar(ECE) for testing the project