--- title: 'Kernel Engineer (Internship and Full-time) at Tilde Research' canonical: 'https://feeny.ai/job/kernel-engineer-internship-and-full-time-tilde-research-san-francisco-w9dnhry0v771' type: 'job' last_seen: '2026-09-06' --- # Kernel Engineer (Internship and Full-time) at Tilde Research - **Company:** Tilde Research - **Location:** San Francisco, CA - **Employment:** full-time - **Work type:** onsite - **Posted:** 2025-07-14 - **Last confirmed live:** 2026-09-06 - **Apply:** https://jobs.ashbyhq.com/tilderesearch/bc4e4071-cf64-4460-8265-b1e5a603d6b8 ## Job description Tilde Research is a moonshot AI lab advancing mechanistic interpretability, new architectures, and pretraining science. We build foundational understanding of models to advance the frontier of intelligence. About the role: As a Kernel Engineer at Tilde, you'll design, implement, and optimize high-performance GPU kernels that are critical to scaling our training and inference workloads. Your work will enable faster iteration cycles, higher throughput, and lower latency. You'll work closely with ML researchers and engineers to co-design models and infrastructure that are deeply performance-aware, and help push the limits of what current hardware can support. What you might work on: - Design, develop, and tune custom GPU kernels for core model operations - Work with ML engineers to prototype and scale novel model architectures - Contribute to system-wide efforts to improve efficiency and throughput, beyond just kernel-level optimizations You're a good fit if you: - Have experience in deep learning or related research areas - Have demonstrated exceptional capability in working on ML kernels. This can include: - Strong open source contributions - Thoughtful technical blog posts/work logs - Previous experience working on hardware-aligned algorithms - Deep familiarity with PyTorch, Triton/TK/TileLang (>1 of), basic familiarity with CUDA, and knowledge of GPU architecture. - Communicate clearly and effectively, both verbally and in writing - Strong algorithmic thinker - Are able to learn quickly ## About Tilde Research ## Company Overview - **One-liner**: Tilde Research is a moonshot AI lab advancing mechanistic interpretability, new architectures, and pretraining science. - **Entity Type**: Private (Seed stage – $8M raised) - **Headquarters**: San Francisco, CA, United States - **Founded**: 2024 - **Founders**: Dhruv Pai (Co-Founder, Chief Science Officer) and Mason Wang (Co-Founder) ## Core Business - **Primary industry**: Artificial Intelligence research – technology, information, and internet. - **Target customers**: Researchers and engineers in the AI/ML community (B2B/Developer‑facing); also seeks to push foundational research that benefits the broader field. - **Mission / purpose**: “To deeply understand the inner mechanisms of models in order to advance the frontier of intelligence.” The lab believes mechanistic insights provide a guiding heuristic for better model design. ## Products & Services The lab publishes open‑source research and tools, primarily on GitHub, including: - **Aurora optimizer** – a new optimizer for training deep models. - **NSA (Native Sparse Attention) kernel** – an efficient implementation of sparse attention. - **MoMoE** – a memory‑optimized Mixture‑of‑Experts implementation for training frontier‑scale MoE models. - **Wall‑Attention** – an attention variant with per‑channel multiplicative decay. - **Activault** – an engine for collecting, uploading, and downloading model activations. - **Sieve** – a tool for applying Sparse Autoencoders (SAEs) to achieve fine‑grained control over model behavior. All products are released as open‑source research artifacts (code and kernels) aimed at advancing interpretability and efficiency. ## Market Standing - **Valuation / Market Cap**: Not publicly available (private company). - **Key Metric**: Total funding of $8M in a seed round (early 2025). - **Notable Investors**: Led by **Khosla Ventures**, with participation from **Forgepoint Capital**, **Soma Capital**, **Timeless**, and **Bain Capital Ventures**. - **Growth Signals**: - Employee headcount: 6 (up 225% YoY). - LinkedIn followers: ~974 (monthly growth +6.7%, yearly +309%). - Website traffic: ~11k monthly visits (yearly growth +87%). - Active open‑source repos with stars (e.g., Aurora optimizer 147 stars, NSA kernel 135 stars). ## Competitive Advantages - **Specialized focus on mechanistic interpretability** – a niche that few labs pursue as a primary moonshot, giving Tilde a unique position between frontier model builders and pure research. - **Open‑source philosophy** – builds community trust and adoption while accelerating the field; attracts top talent who value transparency. - **Small, elite team** with strong academic + industry pedigree (MIT CSAIL, Stanford, UC Berkeley, Anthropic, Amazon). - **Founders deeply embedded in the interpretability community** (Druv Pai has a strong publication record; Mason Wang brings engineering depth). ## Strategic Focus - **Build foundational understanding** of model internals (SAEs, attention mechanisms, MoE routing). - **Enable “control tech”** – interpreter models and tools that allow deep reasoning and command over AI systems. - **Contribute to open‑source infrastructure** (kernels, optimizers) to shape how frontier models are trained and deployed. - **Grow the team** – actively hiring ML researchers (internship and full‑time) to expand capacity. ## Why Work Here - **Cutting‑edge research** – work on the hardest problems in AI interpretability and architecture design. - **Small lab culture** (≈6 people) – high ownership, rapid iteration, direct impact on flagship projects. - **Moonshot mentality** – not product‑driven, but driven by the pursuit of understanding; ideal for researchers who value fundamental science. - **Strong network effects** – team members come from top institutions and labs; alumni go on to places like Nous Research, Forgepoint Capital, and Stanford. - **Career development** – offers both internships and full‑time roles; the interview process includes a graph theory problem that showcases the lab’s intellectual rigor. - **Location**: San Francisco (headquarters) – likely in‑person or hybrid given the small team size; specific remote policy is not publicly stated. ## Sources 1. [tilderesearch.com](https://tilderesearch.com/about) 2. [linkedin.com](https://www.linkedin.com/company/tilderesearch) 3. [github.com](https://github.com/tilde-research) 4. [jobs.ashbyhq.com](https://jobs.ashbyhq.com/tilderesearch/e850d5b0-a5d6-4b9c-9898-f7addb441508) ## Other roles at Tilde Research - [ML Engineer (Internship and Full-time)](https://feeny.ai/job/ml-engineer-internship-and-full-time-tilde-research-san-francisco-zpzxrhy7xesc) — San Francisco, CA - [ML Researcher (Internship and Full-time)](https://feeny.ai/job/ml-researcher-internship-and-full-time-tilde-research-san-francisco-h6065b7073t9) — San Francisco, CA