Waraqa

Waraqa (ورقة — "paper") is a full Research OS: a multi-agent operating system for academic work. It doesn't generate papers wholesale; it runs the process of science — from literature mining to peer review — through a team of specialized agents. The ambition is not "write my paper" but "run my research pipeline," end to end, on your own documents and data.

The pipeline

• Research agent: finds and evaluates sources, scoring them for relevance and methodological fit.
• Writing agent: drafts sections from grounded citations only.
• Integrity agent: checks every claim against its cited source — no citation, no claim.
• Reviewer panel: a calibrated panel of reviewer personas (theoretical, empirical, applied) that critique the draft the way a real conference would.
• Editor agent: turns the accumulated critique into revisions.

Why multi-agent and not one big prompt

A single "write my paper" prompt optimizes for one pass. An agent team separates concerns: retrieval, drafting, verification, critique. Each agent has a narrow objective and an explicit input/output contract, which makes the pipeline auditable — you can see why a claim was dropped.

The reviewer panel is modeled on real peer review because I've been through enough of it: the strongest feedback comes from conflicting perspectives on the same draft.

Research integrity first

The non-negotiable is the citation grounding layer. Generation without grounded citations is a chatbot; with them, it's a research assistant. That framing guides every design decision in the pipeline — and it's why Waraqa is built like an operating system, with agents as composable processes, not a single prompt. The architecture follows the patterns I documented in Multi-Agent Systems: The Patterns That Actually Ship.

Status

In development under the X-Ecosystem. The pipeline runs end-to-end; calibration of the reviewer personas is the current focus.