Soham Ray

Soham Ray is a Research Engineer at Sierra, focused on building conversational AI agents. He has over four years of experience in the customer service AI space, conducting applied research on augmenting and automating task-oriented dialogue systems. He holds a Master’s degree in Computer Science from Cornell University, where he specialized in AI and NLP.

Publicaciones de Soham Ray

White text "τ-voice" on a blurry, grainy background transitioning from orange to purple.

𝜏-voice: benchmarking real-time voice agents on real-world tasks

𝜏-voice is a benchmark for real-time voice agents on 278 grounded customer-service tasks across retail, airline, and telecom. It pairs deterministic, end-to-end task scoring with realistic, controllable audio — diverse personas, environmental noise, and free-form turn-taking.

1 de mayo de 2026

The text "μ-Bench" on a background of purple sky and green grass.

μ-Bench: an open multilingual transcription benchmark

μ-Bench is an open multilingual transcription benchmark built from customer service phone calls across five locales. It aims to measure transcription providers on the errors that matter in production: not just how closely transcripts match their reference, but whether they preserve the speaker’s intent.

20 de abril de 2026

The text "τ³-Bench" on a blurry, grainy background of green and brown.

𝜏³-Bench: Advancing agent benchmarking to knowledge and voice

𝜏³-Bench is here. We've expanded agent evaluation to two new frontiers: knowledge retrieval and voice.

18 de marzo de 2026

Gradient mixed colors, voice and global icons

Multilingual voice: building agents that speak to everyone

Multilingual voice agents aren’t just a feature — they’re a bridge to better, more inclusive customer experiences.

8 de octubre de 2025

𝜏²-bench: benchmarking agents in collaborative real-world scenarios

𝜏²-bench challenges AI agents not just to reason and act, but to coordinate, guide, and assist a user in achieving a shared objective. This leap from solo operation to co-ownership of a task pushes agents into a much more demanding space. And, critically, it reflects the kinds of tasks AI agents are increasingly being asked to perform in the real world.

10 de junio de 2025

Soham Ray

Publicaciones de Soham Ray

𝜏-voice: benchmarking real-time voice agents on real-world tasks

μ-Bench: an open multilingual transcription benchmark

𝜏³-Bench: Advancing agent benchmarking to knowledge and voice

Multilingual voice: building agents that speak to everyone

𝜏²-bench: benchmarking agents in collaborative real-world scenarios

Descubra lo que Sierra puede hacer por usted

Publicaciones de Soham Ray

𝜏-voice: benchmarking real-time voice agents on real-world tasks

μ-Bench: an open multilingual transcription benchmark

𝜏³-Bench: Advancing agent benchmarking to knowledge and voice

Multilingual voice: building agents that speak to everyone

𝜏²-bench: benchmarking agents in collaborative real-world scenarios