Raahul Singh is a Staff AI Research Engineer at Phaidra developing AI systems for the physical infrastructure supporting large-scale computing. His work on Phaidra Prism helps data-center operators navigate sprawling equipment inventories, cooling systems, alarms, and industrial telemetry without relying on model-generated guesses about critical infrastructure.
Singh studied computer science and engineering at the Indian Institute of Information Technology, Sri City. In 2020, he contributed to SunPy and OpenAstronomy through Google Summer of Code, creating Pythia, an open-source project analyzing solar magnetic active regions. His research on solar-flare forecasting explored whether sunspot complexity could help predict flaring activity.
At Phaidra, Singh helped develop the architecture behind Prism, applying machine learning to another complicated physical environment: industrial facilities whose equipment names and telemetry rarely follow consistent conventions. He is also a named co-inventor on a U.S. patent for industrial process control, granted in 2025, covering AI-generated and locally controlled setpoints constrained by operational requirements.
- Semantic blindness in industrial AI. Nearly identical equipment identifiers can defeat vector search, repetitive generated lists can truncate, and splitting inventories across model calls can produce missing or invented assets. Singh and Vanč Levstik addressed these failure modes in their AI Engineer World’s Fair session.
- Hierarchy-aware query planning. Singh’s architecture represents facilities as nested halls, aisles, racks, and equipment. A language model translates an operator’s request into a structured plan; indexed equipment groups, deterministic lookup, and set operations then identify the actual assets. This separates flexible interpretation from exact retrieval and keeps model context bounded as infrastructure grows.
- Uncertainty-aware scientific machine learning. Singh and Ashutosh Pandey coauthored Starkindler, a 2025 research preprint introducing an uncertainty-aware training objective for astronomical redshift estimation. The approach incorporates observational measurement errors directly into model training, extending Singh’s earlier astronomy work toward more reliable scientific predictions.
