Paper Scanner — Research Department Intelligence Engine

Automated discovery and evaluation of research relevant to CoreMind/AION. UAOP Stage 1 for Research.

When to use

• Weekly research scan — "anything new in multi-agent systems?"
• Before architecture decisions — "what does the research say?"
• Technology evaluation — "is this approach validated?"
• Building a literature review for a specific topic

Pipeline

Step 1: Scan Sources

• arXiv (cs.AI, cs.MA, cs.SE) — last 7/30 days
• Conference proceedings (NeurIPS, ICML, KubeCon, PyCon)
• Tech blogs from relevant companies (Anthropic, OpenAI, Google DeepMind)
• Industry reports (Gartner, a16z, Sequoia)

Step 2: Filter and Score

For each paper/article:

• Relevance to CoreMind priorities (multi-agent, LLM, governance, financial ML)
• Evidence grade (A/B/C per research/rules.md)
• Applicability score (1-10): can we use this in the next 90 days?
• Implementation complexity (S/M/L)

Step 3: Produce Research Brief

Per qualifying paper:

• One-paragraph summary (what they did, what they found)
• Key finding applicable to CoreMind
• Implementation proposal (if applicable)
• Evidence grade + confidence level

Step 4: Self-Improvement Loop

• Track research-to-implementation conversion rate
• Reinforce sources with high conversion
• Adjust scanning keywords based on current priorities

Cadence

WEEKLY:   arXiv scan + tech blog digest
MONTHLY:  Full research brief with top 10 findings
QUARTERLY: Research agenda review based on implementation outcomes

Agents

Agent	Role
deep-research (L2)	Owns research agenda, executes scans
data-analyst (L2)	Evaluates quantitative findings
root-cause-analyst (L5)	Validates methodology quality
learning-guide (L5)	Creates learning paths from findings