Anubuddhi: A Multi-Agent AI System for Designing and Simulating Quantum Optics Experiments

TL;DR

<3-5 sentence high-level summary>

Abstract

We present Anubuddhi, a multi-agent AI system that designs and simulates quantum optics experiments from natural language prompts without requiring specialized programming knowledge. The system composes optical layouts by arranging components from a three-tier toolbox via semantic retrieval, then validates designs through physics simulation with convergent refinement. The architecture combines intent routing, knowledge-augmented generation, and dual-mode validation (QuTiP and FreeSim). We evaluated 13 experiments spanning fundamental optics (Hong-Ou-Mandel interference, Michelson/Mach-Zehnder interferometry, Bell states, delayed-choice quantum eraser), quantum information protocols (BB84 QKD, Franson interferometry, GHZ states, quantum teleportation, hyperentanglement), and advanced technologies (boson sampling, electromagnetically induced transparency, frequency conversion). The system achieves design-simulation alignment scores of 8--9/10, with simulations faithfully modeling intended physics. A critical finding distinguishes structural correctness from quantitative accuracy: high alignment confirms correct physics architecture, while numerical predictions require expert review. Free-form simulation outperformed constrained frameworks for 11/13 experiments, revealing that quantum optics diversity demands flexible mathematical representations. The system democratizes computational experiment design for research and pedagogy, producing strong initial designs users can iteratively refine through conversation.

Figures (16)

• Figure 1: Three-Layer Cognitive Architecture of Aṇubuddhi. The system processes natural language requests through three hierarchical layers mirroring expert experimentalist workflow. Layer 1 uses LLM-based intent routing to distinguish questions from design modifications. Layer 2 implements retrieval-augmented generation lewis2020retrieval through semantic search over learned composites. Layer 3 validates designs via dual-mode simulation with convergent self-refinement madaan2023selfrefine. Approved designs feed back as reusable building blocks, implementing experience-based procedural learning.
• Figure 2: Aṇubuddhi-designed optical table layout for Hong-Ou-Mandel interference experiment. Type-II SPDC in BBO crystal generates photon pairs at 810 nm that are spatially separated by PBS based on orthogonal polarizations. HWPs render photons indistinguishable before the delay stage controls temporal overlap at the 50:50 beam splitter where quantum interference occurs.
• Figure 3: Aṇubuddhi-simulated results for Hong-Ou-Mandel interference showing coincidence counts versus delay stage position. The characteristic dip at zero delay (visibility 0.75, FWHM 437 fs) demonstrates quantum two-photon interference with no classical analog. Maximum coincidences (20,761) occur when photons are temporally distinguishable; minimum (2,983) at perfect temporal overlap shows bosonic bunching where both photons preferentially exit through the same output port. Visibility exceeding 0.5 confirms quantum behavior, with excellent signal quality (SNR = 123, true/accidental ratio = 4,910:1).
• Figure 4: Aṇubuddhi-designed optical table layout for Michelson interferometer. Coherent HeNe laser beam is expanded and split into two perpendicular arms by a 50:50 beam splitter. Each arm terminates at a mirror (M1 fixed, M2 piezo-controlled) and returns for interference at the beam splitter, with the combined beams directed to a screen showing the interference pattern.
• Figure 5: Aṇubuddhi-designed optical table layout for Bell state generator. Type-II SPDC in BBO crystal generates entangled photon pairs at 810 nm with orthogonal polarizations (H,V). Mirrors separate the SPDC cone into two spatial paths, with collection lenses, polarizers at 45°, spectral filters, and coupling optics directing photons to SPAD detectors for coincidence measurements demonstrating Bell state entanglement.