← sources

AI agent guardrails for academic research: What broke in our citation and review pipelines

source: reddit (reddit:AI_Agents)published 2026-11-19 18:56ingested 2026-11-20 02:35

retained text

Effective AI agent guardrails for academic research separate generative reasoning from deterministic verification. When unconstrained agents generate text and review citations without code-level boundaries, hallucinated DOIs slip past model self-checks. Fencing model output with external Python scripts that query scholarly APIs ensures zero fake references enter manuscripts.



In our open-source research toolkit created by Sayan Chowdhury at Kalaris Labs, we structured 281 skills around an outer planning loop and an inner deterministic verification loop.



Architectural lessons from building autonomous research agents:

\- Model self-checking fails: Asking an LLM if a citation is real results in plausible confirmation of fake papers.

\- Two-loop boundary: The LLM plans searches and drafts text, while a local Python standard library script intercepts DOIs.

\- Deterministic gating: If a DOI check against Crossref, OpenAlex, or arXiv fails, the script returns a structured error for self-correction.



Install locally via the command line:

npx skills add KalarisLabs/research-agent-skills --skill citation-verification



How are you separating deterministic checks from generative thinking in your multi-agent loops?

document fetched 2026-11-20 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1y0hlsk/ai_agent_guardrails_for_academic_research_what