Panini-DR improves deep research accuracy using multi-agent semantic workspaces
PANINI-DR: Controlled Deep Research through Generative Semantic Workspaces
Computation and LanguageArtificial IntelligenceInformation Retrieval
Summary
Figuring out complex questions often means breaking them into smaller parts and checking each carefully to avoid mistakes. The authors present PANINI-DR, a system that organizes these steps using different smart agents who ask questions, check answers, and keep track of all evidence in a shared workspace. This approach helps find better answers compared to other similar methods, especially on tough questions. PANINI-DR also keeps a detailed record of why each answer was chosen, helping make sure the results are trustworthy.
What this means in practice
- •For enterprise knowledge teams: Maintain and verify evolving research findings during complex investigations to improve accuracy and traceability of answers.
- •For automated helpdesk developers: Build multi-step question answering systems that track dependencies and verify evidence to reduce errors in support responses.
Authors
M. Yigit Turali, Shreyas Rajesh, M. Efe Lorasdagi, Chenda Duan, Kartik Sharma, Vwani Roychowdhury
Abstract
Deep research requires connecting multiple steps of search, reasoning, and computation, with intermediate findings guiding what to investigate next. As investigations grow, agents must maintain research state across model calls and check findings before errors propagate into subsequent work. We introduce PANINI-DR, a multi-agent harness that separates coordination of the overall investigation from the research and verification of individual subquestions. A shared workspace, built on the Generative Semantic Workspace representation, records subquestions, their dependencies, findings, and supporting evidence. Researchers investigate individual subquestions, while a separate verification agent checks proposed answers against those subquestions and their evidence. Verification feedback guides further investigation, and recorded findings inform subsequent assignments. We evaluate PANINI-DR on BrowseComp-Plus, FRAMES, and MoNaCo against seven state-of-the-art deep-research harnesses, all using a shared language-model backbone and frozen corpora. PANINI-DR achieves the strongest average performance across the three benchmarks. On BrowseComp-Plus, it achieves 73.3% accuracy, outperforming the next-best harness by 16.0 points. Its advantage is strongest on hard questions, where it answers nearly twice as many as the next-best harness. Ablations further show that structured planning and persistent state improve performance, as does checking each finding against its cited evidence before it is reused. Because a model can answer correctly from memory alone, we also analyze the evidence retained by each harness. PANINI-DR achieves the highest answer recoverability from saved evidence on all three benchmarks, while complementary claim-support audits show stronger evidence-backed answers. Accounting for this evidence changes how harnesses compare and further strengthens PANINI-DR's advantage.