Saurabh Srivastava

News

• 2025
  Released Rnj-1 (blog, model card), an open-weights 8B-scale code model, with Ashish Vaswani and the team at Essential AI.
• 2024
  Paper: Functional benchmarks for robust evaluation of reasoning performance and the reasoning gap. The framework was subsequently used by Google to evaluate Gemini.
• 2020
  POPL'10 paper "From Program Verification to Program Synthesis" received the SIGPLAN Most Influential Paper award for the prior decade.

Research interests

Code synthesis. LLMs for code and reasoning: data and evals, pre/post training, architectures. Code synthesis from demonstrations and from specifications. Computational synthetic biology. Constraint-based code verification and analysis. Domain-specific languages and compilers.

Highlights

• Essential AI, 2026
  SOTA in its weight class OSS code model.
• Consequent AI, 2024
  Contamination-resistant evals: functional benchmarks.
• Synthetic Minds, 2020
  Automations from screen recordings.
• 20n, 2017
  Design synthetic "acetaminophen yeast" using code synthesis.
• Postdoc, 2014
  Synthesize code that explains biological phenomenon.
• PhD, 2010
  Software that writes code from first principles.

Positions

• 2024 – present
  Head of Research, Code

  Essential AI · San Francisco

  Lead the code research program. Built Rnj-1, an open-weights 8B code model, with Ashish Vaswani. The work spans pre-, mid-, and post-training: multi-trillion-token data construction, stage-specific eval design, agentic environments and instruct/agentic data, and daily ablations on data, architecture, and optimization choices to make outcomes at larger scale predictable from smaller-scale runs. Particular interest in "code worlds" — training data that lets the model reason about idealized behavior (e.g., how it will execute) in addition to the syntactic code.
• 2023 – 2024
  Founder & CEO

  Consequent AI

  Eval frameworks for code and reasoning, and synthetic code datasets for training LLMs. The reasoning-evaluation framework (arxiv:2402.19450) was used by Google to evaluate Gemini.
• 2017 – 2023
  Founder & CEO

  Synthetic Minds (dba Warpdrive) · YC S'18

  Agents for desktop use, synthesized from a small number of recorded demonstrations. Designed a domain-specific language for desktop and browser actions, and a compiler that lifted recordings into programs in this DSL; the synthesizer generalized from three demonstrations to a portable automation. Local background processes interfaced with Windows APIs for low-latency recording and playback. Raised $5.5M, led by Khosla Ventures. Team of 8.
• 2014 – 2017
  Founder & CEO

  20n · YC W'15

  Code synthesis applied to synthetic biology. We engineered a yeast to produce acetaminophen from sugar; previously, acetaminophen was a petroleum derivative. Open-sourced 20n/act, a platform for predicting DNA edits by modeling cell metabolism as programs. Built 40+ organisms for DARPA and two Fortune 500 companies. Raised $2.2M, led by Khosla Ventures. Team of 12.
• 2010 – 2014
  Postdoc and Assistant Research Engineer

  UC Berkeley

  Code synthesis applied to computational and synthetic biology. Built a group of 15 across CS and synthetic biology to develop software that predicted functional genetic edits — for example, modifications of fermentable organisms to produce therapeutics and industrial materials. Started a collaboration with MSR Cambridge on synthesizing biological models from lab observations. Continued work on provably-correct program synthesis, including synthesizers that produce one element of a program pair (e.g., compressor/decompressor, encoder/decoder) given the other.
• 2003 – 2010
  PhD in Computer Science

  University of Maryland, College Park

  Thesis: Satisfiability-based Program Reasoning and Program Synthesis. Showed that if a program can be verified algorithmically, it can also be synthesized algorithmically. Internships at Microsoft Research (Bangalore '08, Redmond '07) and VMware ('04, '05).

Publications

A full list, with citation counts, is on Google Scholar.

• 2024
  Functional benchmarks for robust evaluation of reasoning performance, and the reasoning gap

  Saurabh Srivastava, Annarose M. B., Anto P. V., Shashank Menon, Ajay Sukumar, Adwaith Samod T., Alan Philipose, Stevin Prince, Sooraj Thomas

  arXiv:2402.19450 · 2024

  Subsequently used by Google for Gemini evaluation.
• 2013
  Synthesis of Biological Models from Mutation Experiments

  Ali Sinan Köksal, Yewen Pu, Saurabh Srivastava, Rastislav Bodík, Jasmin Fisher, Nir Piterman

  POPL '13 — 40th ACM SIGPLAN-SIGACT Symposium on Principles of Programming Languages

  acm pdf
• 2012
  Template-based Program Verification and Program Synthesis

  Saurabh Srivastava, Sumit Gulwani, Jeffrey S. Foster

  STTT '12 — Software Tools for Technology Transfer

  springer pdf
• 2011
  Synthesis of First-Order Dynamic Programming Algorithms

  Yewen Pu, Rastislav Bodik, Saurabh Srivastava

  OOPSLA '11 — Object-Oriented Programming, Systems, Languages, and Applications

  acm pdf
• 2011
  Parallel Programming with Inductive Synthesis

  Shaon Barman, Rastislav Bodik, Sagar Jain, Yewen Pu, Saurabh Srivastava, Nicholas Tung

  HotPar '11 — Hot Topics in Parallelism

  acm pdf
• 2011
  Path-based Inductive Synthesis for Program Inversion

  Saurabh Srivastava, Sumit Gulwani, Swarat Chaudhuri, Jeffrey S. Foster

  PLDI '11 — Programming Language Design and Implementation

  acm pdf
• 2010
  From Program Verification to Program Synthesis

  Saurabh Srivastava, Sumit Gulwani, Jeffrey S. Foster

  POPL '10 — Principles of Programming Languages

  acm pdf

  SIGPLAN Most Influential Paper, 2010–2020.
• 2010
  Satisfiability-based Program Reasoning and Program Synthesis

  Saurabh Srivastava

  PhD Dissertation, University of Maryland, College Park

  pdf
• 2009
  VS3: SMT Solvers for Program Verification

  Saurabh Srivastava, Sumit Gulwani, Jeffrey S. Foster

  CAV '09 — Computer Aided Verification (tools paper)

  pdf
• 2009
  Program Verification using Templates over Predicate Abstraction

  Saurabh Srivastava, Sumit Gulwani

  PLDI '09 — Programming Language Design and Implementation

  acm pdf
• 2009
  Constraint-based Invariant Inference over Predicate Abstraction

  Sumit Gulwani, Saurabh Srivastava, Ramarathnam Venkatesan

  VMCAI '09 — Verification, Model Checking, and Abstract Interpretation

  acm pdf
• 2008
  Program Analysis as Constraint Solving

  Sumit Gulwani, Saurabh Srivastava, Ramarathnam Venkatesan

  PLDI '08 — Programming Language Design and Implementation

  acm pdf
• 2008
  Modular Information Hiding and Type Safe Linking for C

  Saurabh Srivastava, Mike Hicks, Jeffrey S. Foster, Patrick Jenkins

  TSE '08 — IEEE Transactions on Software Engineering

  acm pdf
• 2007
  Modular Information Hiding and Type Safe Linking for C

  Saurabh Srivastava, Mike Hicks, Jeffrey S. Foster

  TLDI '07 — Types in Language Design and Implementation

  acm pdf
• 2003
  Distributed Algorithms for Finding and Maintaining a k-Tree Core in a Dynamic Network

  Saurabh Srivastava, R. K. Ghosh

  IPL '03 — Information Processing Letters

  acm pdf
• 2003
  A Code Allocation Protocol for Maximizing Throughput in CDMA-based Ad-hoc Networks

  Saurabh Srivastava, S. Tripathi, A. K. Chaturvedi, D. Sanghi

  WCNC '03 — IEEE Wireless Communications and Networking Conference
• 2002
  Cluster-based Routing using a k-Tree Core Backbone for Mobile Ad-hoc Networks

  Saurabh Srivastava, R. K. Ghosh

  DialM '02 — Workshop on Discrete Algorithms and Methods for Mobile Computing
• 2002
  Approximating the Range Sum of a Graph on a CREW PRAM

  Saurabh Srivastava, P. Gupta

  IWDC '02 — International Workshop on Distributed Computing
• 2002
  Performance Evaluation of Combining Techniques for a Multicarrier CDMA System

  S. Tripathi, Saurabh Srivastava, A. K. Chaturvedi, D. Sanghi

  NCC '02 — National Conference on Communications

Patents

• US 11,966,724 B1
  Building Assisted Automations from Recordings using Program Synthesis

  Granted · Apr 2024
• US 8,402,439 B2
  Program Analysis as Constraint Solving

  Granted · Mar 2013
• US 8,271,404 B2
  Template-based Approach to Discovering Disjunctive and Quantified Invariants over Predicate Abstraction

  Granted · Sep 2012
• US 2017/0260555 A1
  Method for Biosynthesis of Acetaminophen

  Pending · Sep 2017
• US 2017/0211105 A1
  Biosynthetic Production of Carnosine and Beta-Alanine

  Pending · Jul 2017
• US 2017/0211104 A1
  Biosynthetic Production of Acetaminophen, p-Aminophenol, and p-Aminobenzoic Acid

  Pending · Jul 2017
• US 2017/0211103 A1
  Biosynthetic Production of Choline, Ethanolamine, Phosphoethanolamine, and Phosphocholine

  Pending · Jul 2017

Awards

• 2020
  SIGPLAN Most Influential Paper Award (POPL 2010–2020 decade), for "From Program Verification to Program Synthesis."
• 2015
  TechCrunch Top Startups, Y Combinator Winter 2015 batch (honorable mention).
• 2010 – 2012
  ACM-CRA Computing Innovation Fellow.
• 2008 – 2009
  Dean's Fellowship Award in Computer Science (best graduate student), University of Maryland.
• 2002
  Best Undergraduate Thesis in Computer Science, IIT Kanpur.

Education

• 2010 – 2012
  Postdoctoral Researcher, Computer Science

  UC Berkeley

  Code synthesis applied to synthetic biology.
• 2003 – 2010
  PhD, Computer Science

  University of Maryland, College Park

  Thesis: Satisfiability-based Program Reasoning and Program Synthesis.
• 1998 – 2002
  B.Tech., Computer Science

  Indian Institute of Technology, Kanpur

  Thesis: Resource Allocation in CDMA-based Ad-hoc Networks.

Code

• 20n/act — Platform for computational synthetic biology.
• VS3 — Program synthesis over linear arithmetic.
• PINS — Program synthesis using symbolic traces (sources, binaries, inverse mining).

Press

• Sep 2016
  Hard Tech Startups — Y Combinator
• Oct 2015
  DARPA works with six organizations on biological materials for military applications — Military Aerospace
• Mar 2015
  The Top 10 Startups of Y Combinator Winter '15 Demo Day 2 — TechCrunch
• Mar 2015
  Software is designing useful microbes that don't exist in nature — Motherboard
• Feb 2015
  20n, a YC synthetic biology startup, uses software to engineer microbes — TechCrunch