Pankaj Gupta is a co-founder of Baseten, which runs inference for open and custom AI models. Gupta's posts on the Baseten blog cover the company's model performance work, including its adoption of TensorRT-LLM, speculative decoding and inference benchmarks on NVIDIA GH200, and later work on kernel-level optimization for image generation models.1234
At a glance
- Current role
- Co-founder
- Company
- Baseten
- Sector
- Company
- Fields
- Inference and serving
- Work listed
- 1
- Sources cited
- 4
- Profile reviewed
Model performance at Baseten
Gupta is listed on Baseten's blog as a co-founder. In January 2025, Gupta wrote a review of the company's 2024 model performance work, which described adopting TensorRT-LLM as the core framework, using NVIDIA's Hopper architecture and supporting production-ready speculative decoding.12
With Philip Kiely, Gupta tested Llama 3.3 70B inference on a single NVIDIA GH200 in Lambda Cloud (February 2025). In August 2026, Gupta co-authored a post with Brian Li and Faraz Shahsavan on Baseten's agentic kernel optimization framework for image models such as Qwen-Image and FLUX.2.34
Work and contributions
Baseten model performance optimization Proprietary product
Author of Baseten's review of its 2024 model performance work, covering TensorRT-LLM as a core framework, NVIDIA Hopper GPUs and production speculative decoding.2
Timeline
Where they work
- Baseten 3 people in the wiki
Related people
- Tuhin SrivastavaAlso at Baseten
- Amir HaghighatAlso at Baseten
- Pedro CuencaInference and serving
- Furu WeiInference and serving
- Simon MoInference and serving
- Eric HartfordInference and serving
- Patrick von PlatenInference and serving
- Rahul PatilInference and serving
- Sid ShethInference and serving
- Mark SaroufimInference and serving
Suggest a correction or removal
Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.
References
4 sources · profile reviewed Oct 22, 2026 · data updated Oct 22, 2026