I built this to track AI/ML research trends
Future Scan analyzes 187,837 arXiv® papers to identify research themes, gaps, and emerging trends. I use it to support my writing and consulting work.
As a writer and consultant covering AI, I needed a way to spot patterns across thousands of papers. Future Scan analyzes 187,837 arXiv® papers to identify research themes, gaps, and emerging trends.
Currently tracking 187,837 papers | Papers updated every Sunday
Three ways to analyze the data
I designed different analysis types for different questions
Map Research Themes
Shows what research themes exist and how papers cluster around similar concepts.
When I need: A starting point for exploring a new area
Find Research Gaps
Identifies which areas are well-covered versus understudied, revealing research gaps and opportunities.
When I need: To find what's missing or unexplored
Track the Frontier
Focuses on cutting-edge developments and recent advances at the research frontier.
When I need: To stay current with fast-moving areas
Extracting structured data
I added this feature for systematic reviews that need comparable data across papers
How it works
You define custom fields (methodology, sample size, key findings, whatever you need). The AI reads each paper and extracts the information, with confidence scores and supporting quotes so you can verify its work.
Everything exports to CSV for analysis in Excel, R, or your statistical software of choice.
Common use cases
Systematic reviews
Extract standardized information for comparison across studies
Meta-analyses
Collect quantitative results and study characteristics
Literature surveys
Track methodologies or technical approaches across papers
The problem I was trying to solve
I looked at existing tools first, but they didn't solve the specific problem I had.
What I tried
Semantic Scholar, Research Rabbit, Connected Papers: Discovering papers through citation networks and semantic search when you already know where to start.
Elicit, Consensus: Extracting data and answering specific research questions across papers.
Litmaps: Visualizing citation relationships to show how papers connect.
What I still needed
These tools are good at finding specific papers or answering targeted questions. But I needed to see the overall structure of a research area: how papers cluster into natural themes, which areas are crowded versus sparse, and what's emerging versus declining.
So I built this to automatically group papers by semantic similarity (using cosine similarity on embeddings), then analyze each cluster for patterns, gaps, and trends. The goal was to see the forest, not just the trees.
Product demo
Watch a 5-minute video of Future Scan in action
Discovery & organization tools
These tools help me find papers, organize them into collections, and share results.
Semantic Paper Search
Finds relevant papers in ~1 minute with AI-powered semantic search. Provides instant AI summaries and relevance scores without running a full analysis. Includes filtering by author, sorting by relevance or date, and bulk selection.
View sample results →Find Similar Papers
Uses semantic similarity to discover papers related to a specific source paper. Each result includes a similarity score and AI-generated summary explaining the relationship. Results can be shared publicly with a link.
Paper Collections
Organize papers into custom collections (up to 100 papers each). Add personal notes, generate AI summaries of the entire collection, and export to PDF. Use bulk selection to add or remove multiple papers at once. Share collections publicly or keep them private.
Available across all tools:
Bulk Selection
Select multiple papers for batch operations
Advanced Filtering
Filter by author, sort by date or relevance
PDF Export
Download professional reports with summaries
Public Sharing
Share searches and collections with a link
Data source
Future Scan analyzes recent AI/ML research from arXiv, the leading open-access repository for scientific papers.
Covered categories
How it works
Three steps to get results
Enter a research question
Examples: "Security risks of agentic AI" or "Healthcare applications of LLMs"
Pick what you need
Deep Analysis: Research gaps and emerging trends
Data Extraction: Structured data for systematic reviews
Semantic Search: Quick paper discovery with AI summaries
Get results
Deep analysis gives you a thematic report. Data extraction gives you structured tables. Search gives you ranked papers with AI summaries.
What I'm working on next
Features I'm planning to add
Track trends over time
Show how research themes evolve — identify what's emerging vs. declining
Scheduled analysis and email updates
Automatically monitor research trends and send weekly summaries
Cross-domain connections
Identify how ideas and techniques move between different AI/ML subfields
I'm a writer and content strategist who learned systems design, machine learning, and data engineering by building my own tools. I built Future Scan because I needed a way to track AI/ML research trends for my writing.
Karen Spinner, Founder
Learn more about the projectCommon questions
How the features work and what to expect
Trend Analysis
What types of trend analysis are available?
Three types of analysis, each answering different questions:
- Thematic Mapping answers "What are the main topics?" (balanced overview)
- Scoping Review answers "Where are the research gaps?" (identifies what's missing)
- State of the Art answers "What's cutting edge?" (latest breakthroughs)
How does trend analysis work?
Trend analysis uses paper abstracts from arXiv to discover research themes. The process:
- 1. Convert abstracts to math: OpenAI's embedding model converts each abstract to a 1536-dimensional vector that captures its meaning
- 2. Group similar papers: Papers about similar topics naturally cluster together based on mathematical similarity (cosine distance)
- 3. AI reads and summarizes: Claude reads abstracts from the 20 most representative papers in each cluster and writes a summary based on the review type you chose
The results include an interactive map showing research themes and how they relate to each other.
How accurate is trend analysis?
The analysis is based on real papers. Every cited paper is linked to arXiv for verification. Papers are grouped mathematically (not randomly), and AI analyzes abstracts from the 20 papers most central to each cluster. Summaries are based only on the abstracts. Learn more about the methodology →
Data Extraction
What is data extraction?
Data extraction defines custom fields and extracts specific information from paper abstracts. Instead of discovering themes, it collects structured data like methodologies, datasets, performance metrics, or other information.
Useful for systematic reviews and meta-analyses that require comparable data across papers. Results include confidence scores and supporting quotes, and export to CSV for analysis.
How does data extraction work?
A simple five-step process:
- 1. Search for papers: Enter your research question to find relevant papers
- 2. Define fields: Specify what data to extract (methodology, dataset, key findings, etc.)
- 3. AI reads abstracts: Claude reads each paper's abstract and extracts the information you requested
- 4. Review results: Check confidence scores and supporting quotes to verify accuracy
- 5. Export to CSV: Download structured data for analysis in Excel, R, or statistical software
How accurate is data extraction?
Each extracted field includes a confidence score (high, medium, low) and a supporting quote from the paper for verification.
Data extraction analyzes the full PDF content of each paper, not just abstracts. Not all papers contain all requested information—if a field isn't found, the AI indicates it's not found. Learn more about limitations →
Semantic Search
How does semantic search work?
Unlike keyword search, semantic search understands the meaning of the query, not just exact word matches. The process:
- 1. Convert query to vector: OpenAI's embedding model converts the research question to a 1536-dimensional vector
- 2. Find similar papers: The database finds papers whose abstract embeddings are mathematically closest to the query using cosine similarity
- 3. AI summarizes results: Claude reads each paper's abstract and generates a brief summary highlighting its relevance
This allows searching for concepts, not just keywords. A query like "hallucination in language models" produces relevant results even if papers use different terminology.
What papers are in the database?
Future Scan indexes papers from arXiv, an open repository of scientific papers. The database currently contains papers from October 2024 to August 2026 in these categories:
- cs.AI: Artificial Intelligence
- cs.CL: Computation and Language (NLP)
- cs.CV: Computer Vision
- cs.LG: Machine Learning
All analyses (trend analysis, data extraction, semantic search) use paper abstracts—the summaries written by authors that capture key ideas, methods, and findings.
Discovery Tools
How do Collections work?
Collections organize papers into custom groups (up to 100 papers per collection):
- Create collections from search results, similar papers, or cluster themes
- Add notes to individual papers within a collection
- Generate AI summaries of the entire collection showing key themes and connections
- Export to PDF with formatted summaries, notes, and full paper details
- Share publicly via a link, or keep private
What is "Find Similar Papers"?
Using semantic similarity (the same mathematical technique used for clustering), Future Scan finds papers with similar concepts and methods to a source paper. Each result shows a similarity score (0-100%) and an AI-generated summary explaining the relationship. Results can be filtered by author, sorted by similarity or date, and saved to collections.
How do AI summaries work?
AI summaries appear in different contexts:
- Search results: Each paper gets a brief overview of its key contributions
- Similar papers: Summaries explain how each paper relates to the source
- Collection summaries: AI reads all papers in a collection and identifies key themes, connections, and research directions
All summaries are generated by Claude (Anthropic's AI) based solely on paper abstracts. Summaries provide quick understanding without reading full papers.
Can I share results and collections?
Yes. Results and collections can be shared publicly:
- Collections: Toggle sharing on to generate a unique URL. Anyone with the link can view the collection (read-only), including AI summaries and notes
- Similar paper searches: Share similarity search results the same way
- PDF exports: Download and share PDF reports of collections or similarity results
Sharing can be toggled off anytime to make collections private again. Shared items don't require sign-in to view.