Research Atlas
Creator & Lead Developer
Public methods platform
Built a research-methods site and citation-verification skill that cross-checks references against scholarly databases.
Overview
AI-generated reference lists frequently contain fabricated or mismatched citations, which is a real risk for research integrity. Research Atlas addresses that gap. It is a dual project: a public research-methods site at researchatlas.info, plus a research-verification Agent Skill that resolves citations against live scholarly databases instead of trusting model memory. The goal is to help researchers move quickly while keeping provenance and verifiability intact.
My contribution
I designed and built the project end to end: the verification skill, the information architecture and content model, and the frontend. I wrote the verification logic and methodological content, and implemented a static, content-first delivery so updates ship without backend complexity.
What I Built
- A research-verification Agent Skill built on the VERIFY framework that resolves DOI-backed references against CrossRef (primary) and OpenAlex (fallback).
- Title, first-author, and year cross-checks that catch fabricated or mismatched DOIs, with a structured per-reference verdict envelope for each citation.
- A static-first research-methods site (researchatlas.info) organizing reusable prompts, stage-based workflows, and method guides across the research lifecycle.
- Drift-checking and validation in CI so the skill's behavior stays stable as content and rules evolve.
- Multi-agent targeting so the skill runs across Claude Code, Codex CLI, and Gemini CLI.
Why It Matters
Most AI tooling in research optimizes for speed and under-specifies reliability. Research Atlas closes that gap by grounding citation checks in live resolver passes rather than model recall, and by pairing practical workflows with explicit verification rules. The result is a resource researchers can use for daily execution, teaching, and lab standardization.
Technologies & Tools
Deliverables & outcomes
- Published a live research-methods platform at researchatlas.info.
- Built a research-verification skill (v2.1.0) that grounds citation checks in CrossRef and OpenAlex rather than model memory.
- Established a structured per-reference verdict envelope and a locked acceptance corpus to test the skill's reliability.
- Targeted multiple agent runtimes (Claude Code, Codex CLI, Gemini CLI) from a single skill.
- Created a foundation for a broader library of research skills (literature review, academic writing, reproducibility).