Research Paper
Hypothesis Atlas: An Open-Source Evidence Mapping Platform for Multi-Domain Scientific Research
Luxin Zhang · AI Builder & Independent Researcher · United States
Abstract
Hypothesis Atlas is an open-source, web-based research intelligence platform that automatically discovers, analyzes, and visualizes the evidence landscape surrounding a scientific hypothesis. Given a research topic, the system executes an 8-stage asynchronous pipeline that retrieves peer-reviewed papers from OpenAlex and PubMed, experimental datasets from NCBI GEO, and early community signals, then synthesizes results into an interactive knowledge tree that explicitly surfaces supporting evidence, contradicting findings, methodological disagreements, and open research problems.
A built-in Notebook Copilot grounds AI-assisted ideation in real citations. Unlike general-purpose LLMs that synthesize from training data, Hypothesis Atlas pulls from live databases, labels every source by epistemic reliability tier, and treats disagreements and knowledge gaps as first-class outputs. The platform currently supports biotech, technical finance, and chemistry domains, with a roadmap toward fully user-defined topic landscapes.
1. Introduction
The early orientation phase of research — determining where a new hypothesis stands relative to existing literature — is among the most time-consuming aspects of scientific work. Researchers must manually search multiple databases, synthesize conflicting findings, locate relevant datasets, and construct a coherent picture of what is known, contested, and unexplored.
Existing tools each solve only part of the workflow. Citation graph tools map neighborhoods of known work, semantic search tools return relevant papers, and general-purpose LLMs provide fluent summaries. But none consistently combine live retrieval, explicit disagreement surfacing, dataset integration, and epistemic weighting in one hypothesis-first system.
Hypothesis Atlas addresses this with three commitments: hypothesis- first orientation, live multi-source retrieval, and epistemic guardrails that preserve uncertainty rather than hiding it.
2. System Design
2.1 Architecture
Hypothesis Atlas is implemented with Next.js 14 App Router, PostgreSQL via Prisma ORM, and BullMQ + Redis for background processing. The architecture separates API, data, and asynchronous worker responsibilities for reliability and scalability.
2.2 8-Stage Evidence Pipeline
- Initialization and job orchestration
- Query expansion via LLM-generated related keywords
- Paper retrieval from OpenAlex and PubMed
- Dataset retrieval from NCBI GEO
- Social signal retrieval
- Source ranking and deduplication by reliability + recency
- Root knowledge node synthesis
- Child node generation with disagreements and open problems
Pipeline progress is streamed to users in real time using Server-Sent Events (SSE).
2.3 Epistemic Guardrails
Every source is labeled by a reliability tier (peer_reviewed > preprint > dataset > social_signal). Prompts explicitly weight strong evidence and preserve unresolved contradictions, turning disagreement into a first-class signal for future research.
3. Differentiation
Compared with existing tools, Hypothesis Atlas combines live multi-source retrieval, dataset integration, disagreement surfacing, gap-node generation, and notebook-grounded AI ideation in one open-source workflow.
4. Current Domains and Roadmap
Current domains include biotech/life sciences, technical finance, and chemistry. Planned roadmap items include fully user-defined topic seeds, vector retrieval, richer social integrations, PDF full-text extraction, and citation graph visualizations.
5. Conclusion
Hypothesis Atlas automates the evidence orientation phase before research ideation. By combining live retrieval, epistemic labeling, and notebook support, it helps researchers transform uncertainty and disagreement into structured, actionable research directions.