Luminae

The problem

Researchers spend hours manually critiquing methodology, auditing datasets and drafting grant proposals from academic papers.

What was built

A five-agent research-paper intelligence platform — OCR, vision, critics, synthesis and grant — that critiques methodology, audits datasets, proposes experiments and drafts NSF/NIH-style grant outlines using Mistral models.

Architecture and engineering decisions

A FastAPI service orchestrates five specialised agents over uploaded PDFs, with Mistral OCR and vision models feeding critique and synthesis steps. Budget middleware caps model spend, and a Next.js interface presents results.

Audience

Academic researchers and research teams.

Technical highlights

• ✓Five specialised agents: OCR, vision, critics, synthesis, grant
• ✓PDF upload with OCR and vision analysis
• ✓Methodology critique with severity ratings
• ✓Dataset auditing and bias detection
• ✓NSF/NIH-style grant outline generation
• ✓Budget middleware capping model spend

My responsibilities

• ✓Product design
• ✓Full-stack implementation
• ✓Multi-agent orchestration
• ✓Budget controls

Why it matters for a client

Shows multi-agent orchestration, document processing and cost controls for an AI workflow with real budget constraints.

Verified evidence

Live at luminae.qzz.io and described as built by DevRay Labs. The owner portfolio reports five agents, grant generation and a README-estimated cost of roughly $0.05 per paper with budget middleware.

Limitations

Hackathon-built; accuracy of critiques and grant output are not independently verified; do not claim zero hallucinations.

Technology