# Alex Chen

> Alex (Wei) Chen is a Director at Qualcomm and the founder, former CEO, and Chief Scientist of Nexa AI, a generative AI inference company acquired by Qualcomm in 2026. He is the main technical author of NexaSDK, the Octopus agent models, OmniVLM, OmniAudio, and NexaQuant. This work has been featured by Qualcomm, NVIDIA, AMD, Intel, Google DeepMind, Google, OpenAI, Hugging Face, IBM, Alibaba Qwen, Liquid AI, Microsoft, and Apple. He holds a PhD from Stanford University (2024) and works on AI systems, generative AI inference, hardware-software co-design, and robotics.

This file follows the llms.txt convention (https://llmstxt.org). It summarizes https://alexchen.ai for language models and agents.
A full-text Markdown export of every blog post is at https://alexchen.ai/llms-full.txt.

## Key facts

- Name: Alex Chen (also published as Wei Chen, Alex (Wei) Chen, Wei (Alex) Chen)
- Current role: Director, Qualcomm (https://www.qualcomm.com), Santa Clara, CA
- Founded: Nexa AI (https://nexa.ai), 2024 — Generative AI inference company (NexaSDK, NexaML, Octopus, OmniVLM, OmniAudio, NexaQuant); acquired by Qualcomm in 2026.
- Education: PhD, Stanford University (2024)
- Awards: Best Paper Runner-Up Award, IEEE International Conference on Data Mining (ICDM 2025)
- Recognized by: Qualcomm, NVIDIA, AMD, Intel, Google DeepMind, Google, OpenAI, Hugging Face, IBM, Alibaba, Liquid AI, pyannoteAI, Microsoft, Apple (sources below)
- Areas: AI systems; Generative AI inference; Large language model inference; Vision-language models; Model quantization; Hardware-software co-design; NPU and GPU acceleration; Robotics; Imitation learning; Reinforcement learning for LLMs; GPU kernel design
- Contact: alexchen4ai@gmail.com
- Profiles: https://github.com/alexchen4ai, https://www.linkedin.com/in/wei-chen-stanford, https://x.com/alexchen4ai

## Industry recognition

Third-party coverage of this work at Nexa AI, grouped by sector. Each link is the original source.

### Silicon

- [Qualcomm (Blog)](https://www.qualcomm.com/developer/blog/2025/11/nexa-ai-for-android-simple-way-to-bring-on-device-ai-to-smartphones-with-snapdragon): Developer blogs on Snapdragon Android, the Hexagon NPU, Granite 4.0, and IoT & robotics.
- [Qualcomm (LinkedIn)](https://x.com/Qualcomm/status/1988705450112794704): Featured on Qualcomm's X and LinkedIn, plus seven “This Week in AI” posts (1, 2, 3, 4, 5, 6, 7).
- [NVIDIA](https://blogs.nvidia.com/blog/rtx-ai-garage-nexa-hyperlink-local-agent/): RTX blogs on our local agent and CES booth, a 1.3M-view video.
- [AMD (Blog)](https://www.amd.com/en/developer/resources/technical-articles/2025/advancing-ai-with-nexa-ai--image-generation-on-amd-npu-with-sdxl.html): Blogs on NPU image generation and DeepSeek R1 speedups with NexaQuant.
- [AMD (LinkedIn)](https://www.linkedin.com/posts/amd_hyperlink-stop-searching-start-asking-activity-7371610679142891521-VpD4): Posts from AMD, AMD Enterprise, and AMD Developer, plus an interview on its developer channel.
- [Intel](https://www.linkedin.com/posts/intel-software_ai-ondeviceai-nexasdk-activity-7376337062087667712-xw7i): Spotlighted NexaSDK on its developer channel for on-device LLMs on Intel's NPU.

### Models

- [Google DeepMind](https://deepmind.google/models/gemma/gemmaverse/omniaudio/): Our model OmniAudio is featured by Google DeepMind in its Gemmaverse, a showcase of models built on Gemma.
- [Google (Blog)](https://developers.googleblog.com/en/gemma-family-and-toolkit-expansion-io-2024/): Highlighted in the 2024 Gemma keynote, with posts from Google AI Developers and Google Devs.
- [OpenAI](https://x.com/dkundel/status/1976769779336987069): Its Developer Relations team amplified our work running open models on consumer hardware.
- [Hugging Face](https://x.com/Thom_Wolf/status/1864227910099366327): Ranked our models among the most-downloaded from 2022 to today; CEO Clément Delangue called to collaborate.
- [IBM](https://www.ibm.com/new/announcements/ibm-granite-4-0-hyper-efficient-high-performance-hybrid-models): Listed NexaSDK as a supported framework for Granite 4.0, next to vLLM, llama.cpp, and MLX.
- [Alibaba](https://x.com/Alibaba_Qwen/status/1978154384098754943): The Qwen team called out our day-0 Qwen3-VL support and Qwen3 on the Qualcomm NPU.
- [Liquid AI](https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai): Partnered with us on the LFM2.5 launch and LFM2.5 Thinking, amplified by a Qualcomm VP.
- [pyannoteAI](https://www.linkedin.com/posts/pyannoteai_were-really-excited-to-join-ces-2026-through-activity-7414255667328212992-itTh/): Ran their diarization model on Qualcomm NPUs through NexaSDK for CES 2026.

### Platforms

- [Microsoft](https://blogs.windows.com/windowsexperience/2025/11/18/ignite-2025-windows-at-the-frontier-of-work/): Demoed our product on stage at the Ignite 2025 keynote and listed us as a software partner for Windows.
- [Apple](https://x.com/nexa_ai/status/1955313386746749025): The Apple MLX team reposted our work running multimodal models on Apple silicon.

### Talks

- [Microsoft Ignite 2025](https://blogs.windows.com/windowsexperience/2025/11/18/ignite-2025-windows-at-the-frontier-of-work/)
- [CES 2026 (NVIDIA & AMD booths)](https://blogs.nvidia.com/blog/rtx-ai-garage-ces-2026-open-models-video-generation/)
- [Qualcomm Snapdragon Summit](https://www.qualcomm.com/developer/blog/2026/03/run-nexa-ai-agents-locally-on-snapdragon-pc-with-hexagon-npu)
- [GenAI Summit San Francisco](https://sf2024.genaisummit.ai/#/)
- PyTorch Conference
- AMD Advancing AI Event
- Web Summit
- IBM TechXchange
- Open Data Science Conference

## Experience

- Director (Principal Engineer/Manager), Qualcomm (2026–Present)
- Founder, CEO and Chief Scientist, Nexa AI (2024–2026)
- PhD, Stanford University (2019–2024)

## Selected work

- [NexaSDK](https://github.com/NexaAI/nexa-sdk) (AI Infra): An open-source generative AI inference runtime that runs LLM, vision-language, and audio models across NPU, GPU, and CPU. Hit #1 on GitHub Trending with 8,000+ GitHub stars, and recognized by Qualcomm, AMD, NVIDIA, Intel, IBM, and Microsoft.
- [Hyperlink](https://blogs.nvidia.com/blog/rtx-ai-garage-nexa-hyperlink-local-agent/) (Application): A local AI agent that lives inside your computer — it searches across all your files, fully private. Reached 2.1M views and 30K users in two months, featured by NVIDIA, Microsoft, Qualcomm, and AMD.
- [Octopus and Omni Model Series](https://arxiv.org/abs/2404.01744) (Models): The first AI agent models built to run directly on hardware — function calling across PCs, phones, wearables, automotive, and robots — plus the OmniVLM and OmniAudio multimodal models. The line trended top-3 on Hugging Face and was featured at Google I/O and by Google DeepMind.

## Pages

- [About](https://alexchen.ai/): Bio, latest posts, selected work (NexaSDK, Hyperlink, Octopus and Omni models), selected publications, industry recognition, experience, and FAQ
- [CV](https://alexchen.ai/cv/): Experience, education, awards, and publications
- [Publications](https://alexchen.ai/publications/): Peer-reviewed papers and preprints
- [Recognition](https://alexchen.ai/recognition/): Coverage from Qualcomm, NVIDIA, AMD, Microsoft, Google, IBM, Intel and others
- [Blog](https://alexchen.ai/blog/): Writing on AI systems, LLM inference, and robotics
- [Projects](https://alexchen.ai/projects/): Selected engineering and research projects
- [RSS feed](https://alexchen.ai/feed.xml)
- [Sitemap](https://alexchen.ai/sitemap.xml)

## Blog posts

- [The AI Coding Paradox: When Faster Code Makes Projects Slower](https://alexchen.ai/blog/2026/the-ai-coding-paradox/): Why AI-generated code accelerates some teams and slows others — coding velocity vs. delivery velocity, the rise of AI slop, and a short field guide for using AI without losing the plot. (2026-09-20; tags: ai-coding, software-engineering, productivity, agents)
- [Understanding Kernel Design from a Mathematical Perspective](https://alexchen.ai/blog/2026/kernel-math/): Relating mathematical concepts — index notation, associative mergeable summaries — to GPU kernel design for ML inference, with case studies on fused operators and FlashAttention. (2026-05-11; tags: kernel, inference, optimization)
- [Gated Delta Networks: Improving Mamba2 with the Delta Rule](https://alexchen.ai/blog/2026/gated-delta-networks/): A reading note on Yang et al., 2024 — how combining Mamba2's decay gate with DeltaNet's selective memory update yields a flexible recurrent model, and why its parallel training algorithm requires a matrix inverse. (2026-04-13; tags: transformers, mamba, linear-attention)
- [Universal Manipulation Interface: In-The-Wild Robot Teaching Without In-The-Wild Robots](https://alexchen.ai/blog/2026/universal-manipulation-interface/): A reading note on Chi et al., 2024 — how a handheld gripper with a fisheye camera and mirrors enables hardware-agnostic demonstration collection anywhere, and what retargeting steps bridge the gap to a real robot. (2026-04-05; tags: robotics, imitation-learning, manipulation)
- [Diffusion Policy: Visuomotor Policy Learning via Action Diffusion](https://alexchen.ai/blog/2026/diffusion-policy/): A reading note on Chi et al., 2023 — how DDPM-style denoising is applied to robot action generation, why it handles multimodal demonstrations better than explicit policies, and what the inference cost looks like in practice. (2026-04-04; tags: robotics, diffusion, imitation-learning)
- [Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware](https://alexchen.ai/blog/2026/act-action-chunking-with-transformers/): A reading note on ALOHA (Zhao et al., 2023) — how ACT uses action chunking and a Conditional VAE to solve fine manipulation tasks that cause standard imitation learning to fail completely. (2026-04-04; tags: robotics, transformers, imitation-learning, manipulation)
- [Reinforcement Learning for Large Language Models](https://alexchen.ai/blog/2026/reinforcement-learning-for-llms/): From the fundamentals of RL — value functions, policy gradients, Q-learning, PPO — to how these ideas translate directly into RLHF and DPO for LLM alignment. (2026-02-14; tags: llm, reinforcement-learning, rlhf, dpo, ppo)
- [Optimizing LLM Inference for On-Device Deployment](https://alexchen.ai/blog/2026/llm-inference-optimization/): A practical overview of quantization, pruning, KV cache, and hardware-specific techniques for running large language models faster — with a focus on edge devices. (2026-02-11; tags: llm, inference, optimization, quantization, on-device-ai)
- [Evaluation of Large Language Models](https://alexchen.ai/blog/2026/evaluation-of-large-language-models/): A practical guide to LLM benchmarks — what they measure, how they are computed, and how to run them yourself. (2026-02-11; tags: llm, evaluation, benchmarks)

## Publications

- Wei Chen, Zhiyuan Li, Mingyuan Ma (2025). Octopus: On-device language model for function calling of software APIs. Proceedings of the 2025 Conference of the North American Chapter of the Association for Computational Linguistics: Industry Track. https://arxiv.org/abs/2404.01549
- Wei Chen, Zhiyuan Li, Zhen Guo, Yikang Shen (2025). Octo-planner: On-device Language Model for Planner-Action Agents. Proceedings of the Workshop on Empowering Multi-Agent Systems (EMAS 2025). https://arxiv.org/abs/2406.18082
- Honghui Xu, Shiva Shrestha, Wei Chen, Zhiyuan Li, Zhipeng Cai (2025). DP-FedLoRA: Privacy-Enhanced Federated Fine-Tuning for On-Device Large Language Models. IEEE International Conference on Data Mining (ICDM). Best Paper Runner-Up Award. https://arxiv.org/abs/2509.09097
- Wei Chen, Liangmin Wu, Yunhai Hu, Zhiyuan Li, Zhiyuan Cheng, Yicheng Qian, Lingyue Zhu, Zhipeng Hu, Luoyi Liang, Qiang Tang et al. (2025). AutoNeural: Co-Designing Vision-Language Models for NPU Inference. https://arxiv.org/abs/2512.02924
- Wei Chen, Zhiyuan Li, Shuo Xin (2024). OmniVLM: A Token-Compressed, Sub-Billion-Parameter Vision-Language Model for Efficient On-Device Inference. https://arxiv.org/abs/2412.11475
- Wei Chen, Zhiyuan Li, Shuo Xin, Yihao Wang (2024). Dolphin: Long Context as a New Modality for Energy-Efficient On-Device Language Models. https://arxiv.org/abs/2408.15518
- Wei Chen, Zhiyuan Li (2024). Octopus v4: Graph of language models. https://arxiv.org/abs/2404.19296
- Wei Chen, Zhiyuan Li (2024). Octopus v3: Technical Report for On-device Sub-billion Multimodal AI Agent. https://arxiv.org/abs/2404.11459
- Wei Chen, Zhiyuan Li (2024). Octopus v2: On-device language model for super agent. https://arxiv.org/abs/2404.01744
- Jiajun Xu, Zhiyuan Li, Wei Chen, Qun Wang, Xin Gao, Qi Cai, Ziyuan Ling (2024). On-Device Language Models: A Comprehensive Review. https://arxiv.org/abs/2409.00088

## Optional

- [Full blog text](https://alexchen.ai/llms-full.txt): Every post as Markdown, newest first