Automated AI Research Monitoring

The Problem

Staying current with AI research across multiple domains requires daily monitoring of arXiv and academic databases. Manual checking of 8 research topics with multiple keyword variations is tedious and easy to miss important papers.

The volume is overwhelming. The relevance filtering is manual. The summarization takes time.

The Solution

Automated research monitoring pipeline that handles collection, deduplication, summarization, and notification.

How It Works

Multi-source search — queries arXiv API (XML parsing) and Semantic Scholar API with configurable rate limiting.

Smart deduplication — tracks seen URLs and cross-references arXiv IDs across sources. No duplicate papers in your feed.

AI Summarization — Claude API extracts key insights in digestible format. Read the summary, decide if the full paper is worth your time.

Daily digest — markdown file with all papers organized by topic for offline reading.

Telegram notification — immediate alert when new papers arrive, with links and topic categorization.

Research Topics Monitored

The system tracks 8 configurable research areas:

Continual Learning — catastrophic forgetting, lifelong learning

Memory & Architectures — episodic memory, working memory, memory-augmented networks

Embodiment & Robotics — sensorimotor learning, proprioception, embodied AI

Qualia & Consciousness — machine consciousness, IIT, phenomenal experience

Hybrid Architectures — neuro-symbolic, hybrid neural systems

World Models — JEPA, predictive models, latent world models

Edge AI — efficient inference, on-device ML, compression

Alignment & Safety — reward hacking, value alignment, safe RL

Technical Details

Rate limiting — 500ms delay between arXiv queries, 1s between Semantic Scholar. Respectful API usage.

Category coverage — arXiv categories cs.LG, cs.AI, cs.NE, cs.RO, cs.CL, cs.AR, cs.DC, cs.CY, q-bio.NC

Configurable window — default 7-day lookback, adjustable per topic.

launchd scheduling — runs at 6:00 daily, fully automated.

Workflow

Automatic (daily at 6:00)

arXiv + Semantic Scholar → Dedupe → Summarize → Desktop file + Telegram notification

Manual

npm start to run immediately when you want fresh results.

Results

• 8 research topics monitored daily
• ~30 papers per run (configurable limit)
• 7-day window for catching papers you might have missed
• Zero manual effort after initial setup

Architecture

Clean separation of concerns:

• services/arxiv.js — arXiv API client (XML → JSON)
• services/semantic-scholar.js — Semantic Scholar API client
• services/summarizer.js — Claude API summarization
• services/file-writer.js — daily digest markdown generator
• telegram.js — notification delivery
• config.js — topics, keywords, limits

Future Considerations

• Paper relevance scoring (similar to Content Radar)
• Citation tracking for followed papers
• Integration with Zotero/Notion for paper management
• Clustering related papers across topics

Internal tool for research awareness. Architecture available as reference for monitoring automation projects.