
- Joined
- Sep 2025
- How they joined
- Directly
- Before Cognition
- Jane StreetLinux Engineer
- Status
- Current
- Credited as the second-listed author on Cognition's 'Introducing SWE-1.5: Our Fast Agent Model' (2025-10-29), a frontier-size coding model served via Cerebras at up to 950 tok/s and shipped in Windsurf; he joined Cognition as Member of Technical Staff only about two months earlier (Sep 2025).Oct 29, 2025
- Cognition signed a definitive agreement to acquire Windsurf (the agentic IDE, formerly Codeium) on 2025-07-14, including its IP, product, brand and people; Nikhil joined Cognition in Sep 2025 from Jane Street.Jul 14, 2025
- Wrote Jane Street's 'A Higgs-bugson in the Linux Kernel' (2025-07-02), debugging a hard-to-reproduce kernel bug in Gord, the firm's system that stores and distributes trading activity data, over NFSv3 with Kerberos; his last public Jane Street artifact, published a month before he left.Jul 2, 2025
- Linux Engineer at Jane Street in New York (Aug 2024 - Aug 2025) after a summer 2023 Technology Internship there; Berkeley Traders Association lists him as 'Software Engineer, Jane Street, Class of 2024'.2024
- Co-author of 'NeCTAr: A Heterogeneous RISC-V SoC for Language Model Inference in Intel 16' (arXiv 2025-03-18) with Viansa Schmulbach, Jason Kim, Ethan Gao, Lucy Revina, Ethan Wu and Borivoje Nikolic, a 16nm chip tapeout running sparse ReLU-Llama inference; ties to his EE194 Intel tapeout TA role.Mar 18, 2025
- Undergraduate researcher and systems administrator in Ken Goldberg's AUTOLab (Berkeley AI Research) from Jan 2021 to May 2024: wrote a Kubernetes scheduler for GPU ML jobs, automated permission/hosting requests, and built a system offloading compute-heavy ROS nodes to AWS EC2.Jan 2021
- Led the Berkeley Open Computing Facility (OCF), the all-volunteer student computing org: Site Manager (2020-2021), Deputy Site Manager, then General Manager (May 2022 - May 2023); migrated its infra from Puppet to a bare-metal Kubernetes cluster serving 100k+ users.Jul 2020
- SWE-1.5 co-authors listed with him include Jacob Teo, Connor Fogarty, Gary Chang, Theodor Marcu, Edison Zhang and Albert Tam, his nearest named collaborators at Cognition.Oct 29, 2025
- Built RoSE (Remote Shell Environment), a Mosh replacement in Rust over QUIC using wezterm's terminal emulator (43 stars), hosted at rose.nikhiljha.com.
- Recent repos since joining Cognition: brouter (Swift macOS link router for browser profiles, Sep 2026), backforagammon (Jul 2026), jj-vanity (custom Jujutsu change IDs, Apr 2026) and a polydup fork, a steady side-project cadence while at Cognition. He has also forked SWE-agent, vllm, skypilot and chipyard.Sep 2026
- Other Rust/infra repos: mycelium (Minecraft server networks on Kubernetes), aeneid (GitHub SSH keys for sshd auth), adelie (keeps git-ops deployments up to date), arcanum (Vault secret sync operator); he also wrote ocf/postgres-operator for OCF.2024
- TA at Berkeley (Jun 2022 - May 2024) for CS61C Computer Architecture (two semesters), CS161 Computer Security (two semesters), EECS151 Digital Design, and EE194 Intel 22FFL Tapeout.Jun 2022
TA for CS61C (Computer Architecture) for two semesters, CS161 (Computer Security) for two semesters, EECS151 (Digital Design) for one semester, and then EE194 (Intel 22FFL Tapeout).
•Reduced machine usage conflicts by writing a scheduler for GPU ML/AI jobs on Kubernetes. •Eliminated an order of magnitude of researcher downtime by automating and writing an API for permission and hosting requests,administrative changes, and other services. •Created a system to automatically offload compute-intensive ROS nodes to AWS EC2. •Tools Used: Python 3, Kubernetes, Ansible, Vector, Grafana, ROS
•Served 100k+ end-users with petabytes of data per year as technical lead of the Open Computing Facility. •Migrated our infrastructure from Puppet to a bare metal Kubernetes cluster, built out from scratch with a small team on a low budget. •Protected against 0ld-day attacks by writing a low-maintenance tool to automate dependency and infrastructure updates. •Improved network-wide observability by configuring distributed tracing, logging, and monitoring. •Tools Used: Kubernetes, Rust, Python 3, ElasticSearch, Puppet, ArgoCD, Cilium