Bimal Gaudel
C++ Systems · Numerical Libraries · Tensor Algebra & Scientific Computing
Open to work. Looking for systems and runtime roles, numerical and simulation libraries, scientific software R&D, and HPC research-software / postdoc roles. Based in Blacksburg, VA. Open to relocation.
bimalgaudel@gmail.com Download CV
I led the development of SeQuant, a tensor-algebra framework with a compiler-style pipeline. It takes a symbolic quantum-chemistry derivation and runs it as numerical code at supercomputer scale — work that took researchers months now takes days.
Technical expertise
Modern C++ (17/20/23) — production runtimes, type-driven design, and extensible library APIs. Around it I build the pieces scientific software runs on: distributed-memory parallelism, dynamic optimizers over runtime data, graph-theoretic transforms, and memoization-cache design, all validated at supercomputer scale.
SeQuant
I designed and led the implementation of SeQuant: an embedded DSL, a graph-canonicalized IR, a cost-based optimizer, an interpreter, and a transpiler.
- SeQuant expressions double as the AST in idiomatic C++. Node identity in the IR comes from tensor-network graph canonicalization, which makes common-subexpression elimination work across separate equations.
- Contraction ordering is cost-based — e.g. subset dynamic programming over tensor networks, scored by flop and memory counts. The interpreter evaluates the IR with memoization keyed on canonical identity and memory-aware cache eviction.
- The backend interface is extensible: TiledArray, BTAS and TAPP backends ship with SeQuant, and the IR also lowers to C++/Python for downstream tensor compilers like TACO.
Full architecture described in (Gaudel et al., 2026).
TiledArray
Contributing developer on TiledArray, the massively-parallel block-sparse distributed tensor framework. Extended its operation basis to support tensor-of-tensor data structures, enabling block-wise compression and PNO-like local-correlation methods.
Experience
Lead developer, SeQuant | Research Associate, Virginia Tech 2018 - 2025
Developed a runtime that optimizes symbolic tensor expressions for distributed-memory evaluation in TiledArray.
Education
PhD in Theoretical & Computational Chemistry | Virginia Tech 2018 - 2025
Dissertation: Automated implementation of advanced electronic structure methods
Publications
SeQuant (Gaudel et al., 2026)
A color-graph approach to canonicalizing tensor networks, used for symbolic transformation and runtime evaluation of many-body methods.
Applied research using SeQuant
- Theoretical exploration of new ansatze in explicitly correlated methods (Masteran et al., 2025).
- Geminal parameter tuning in explicitly correlated methods (Powell et al., 2025).
- Discovery of effective theories compared to complex counterparts (Teke et al., 2024).
- Identification and correction of errors in previously published works (Masteran et al., 2023).
Blog posts
Fast RTTI in C++ for a class hierarchy
Compile-time type-id generation for performant runtime type inference across a class hierarchy.
The access-by idiom in C++
A pattern for unit-testing private methods when no better seam exists.