Neel Somani
University of California, Berkeley | Class of 2019
B.A. Mathematics | B.A. Computer Science | B.S. Business Administration
Major GPA: 4.0
Bio
Neel Somani is a machine learning researcher focused on formal methods, mechanistic interpretability, and verification-oriented approaches to neural systems. Neel graduated from UC Berkeley with a triple major in Mathematics, Computer Science, and Business Administration. His contributions to security and privacy have been recognized at top conferences such as USENIX Security and OOPSLA, where Duet received the ACM SIGPLAN Distinguished Paper Award.
Previously, Neel founded Eclipse Labs, where he architected the first Solana Virtual Machine (SVM) Layer 2. He has raised $65M for Eclipse and backed multiple billion-dollar crypto protocols through Chord Ventures. Neel supports education through the Neel Somani Scholarship Program and has served on the Board of the Berkeley-Haas Alumni Network. He has been recognized as an Accel Scholar, Phi Beta Kappa, and Cal Alumni Association Leadership Award recipient.
Research
Research Interests: Formal methods and systems verification. Mechanistic interpretability. Theoretical foundations of neural architectures.
Past: Privacy-preserving ML and applied cryptography. Blockchain and decentralized systems.
Research Experience
- PrivGuard | 2022 | UC Berkeley, Prof. Dawn Song
"PrivGuard: Privacy Regulation Compliance Made Easier"
L Wang, U Khan, J Near, Q Pang, J Subramanian, N Somani, P Gao, et al.
31st USENIX Security Symposium, 2022 | NSF Award #1518899 - Data Capsule | 2019 | UC Berkeley, Prof. Dawn Song
"Data Capsule: A New Paradigm for Automatic Compliance with Data Privacy Regulations"
L Wang, JP Near, N Somani, P Gao, A Low, D Dao, D Song
Heterogeneous Data Management, Polystores, and Analytics for Healthcare, 2019 - Oasis Labs | 2019 | UC Berkeley, Prof. Dawn Song
Privacy-preserving ML using SGX enclaves and blockchain - Duet | 2017–2018 | UC Berkeley, Prof. Dawn Song
"Duet: An Expressive Higher-order Language and Linear Type System for Statically Enforcing Differential Privacy"
JP Near, D Darais, C Abuah, T Stevens, P Gaddamadugu, L Wang, N Somani, et al.
Proceedings of the ACM on Programming Languages (OOPSLA), 2019 | ACM SIGPLAN Distinguished Paper Award - metaCurious | 2019 | Berkeley Institute for Data Science, Prof. JB Poline
"metaCurious: a web-application for meta-analytic datasets curation and annotation"
K Bottenhorn, A Keshavan, N Somani, S Ramesh, JB Poline, A Laird
BRAIN Initiative Investigators Meeting, 2019 - Cirrus | 2018 | UC Berkeley RISELab, Prof. Randy Katz
Serverless machine learning framework | GitHub - Brainspell | 2016–2018 | Berkeley Institute for Data Science, Prof. JB Poline
"Collaborative curation of articles collections for meta-analyses in brain imaging: Brainspell-neo"
N Somani, S Ramesh, A Keshavan, R Toro, JB Poline
Organization for Human Brain Mapping, 2018
Current Projects
- Symbolic Circuit Distillation | 2025 | Formal Equivalence Prover from Circuits to Programs
This project extracts a Python program from a weight-sparse LLM circuit and uses an SMT solver to prove bounded-domain equivalence between them.
GitHub: https://github.com/neelsomani/symbolic-circuit-distillation - Cuq | 2025 | Formal Verifier for Rust CUDA Kernels
A tool that formally verifies safety properties of Rust-based CUDA kernels.
GitHub: https://github.com/neelsomani/cuq - Do Models Encode Epistemic Stance in Reasoning? | 2025 | Causal Analysis
LLMs encode assumed truth/falsity as a low-dimensional, assumption-scoped control signal that causally mediates premise-conditioned inference.
GitHub: https://github.com/neelsomani/epistemic-stance-mechinterp
Selected Honors and Awards
- Accepted to Y Combinator | S22 batch (declined) | 2022
- Phi Beta Kappa | 2019
- Magna Cum Laude | 2019
- Haas Undergraduate Scholarship | 2019
- Cal Alumni Association Leadership Award | 2019
- Kairos Fellow | Kairos Society | 2018
- Accel Scholar | 2018
- EECS Honors Program | 2017
- Upsilon Pi Epsilon | 2017
- Lipson Essay Prize | "Contemporary Stereotyping: Exploring the Seduction of Bias" | 2016
- Valedictorian | Venture High School | 2015
Teaching & Presentations
Teaching
- Head Lecturer | 2018–2019 | UC Berkeley, Prof. Thomas Lee
UGBA 198: Methods and Mathematical Foundations of Machine Learning for Business Decisions
Created and taught the first ML for business course at Haas with 200+ students per semester
Managed team of 9 TAs and graders
Selected Conference Presentations
- From High Fees to High Speeds: Optimizing the SVM For Ethereum | 2024: YouTube
- Challenges & Advantages of Parallel VM Layer-2s | 2024: YouTube
- Is Data Availability (DA) a Good Business Model | 2024: YouTube
- Rollups-as-a-Service Are Going To Zero | 2023: YouTube
- Design Trade-Offs for Persistent App-Specific Rollups | 2023: YouTube
Professional Activities
- Founder: Eclipse Labs - Raised $65M, architect of Solana Virtual Machine Layer 2
- Managing Partner: Chord Ventures - Backed multiple billion-dollar crypto protocols
- Donor: Neel Somani Scholarship Program - Supporting education
- Board Member: Berkeley-Haas Alumni Network, San Francisco Chapter
- Quantitative Researcher: Citadel - Commodities, electricity grid optimization
Job Openings
- Chord Ventures Summer Incubator: Program for high-achieving students in technical majors (CS, Math) with demonstrated software development ability. Applications open in spring.
- Eclipse Labs & Portfolio Companies: Hiring engineering talent. Priority given to candidates with experience at top-tier companies (web2 or web3).
Contact
- Email: ude.yelekreb@leen
- Website: neelsomani.com
- Blog: neelsomaniblog.com
- X/Twitter: @neelsomani
- Google Scholar: Publications