Vaibhav Bajpai

Research Collaboration

I am currently seeking research collaborations on four prototypes. Each project combines artificial intelligence, simulation, or statistical modeling with a Lean 4 verification layer, ensuring that its results are supported by machine-checked evidence that can be independently audited.

Automated theorem proving

QEDAI

From natural-language mathematics to Lean 4–verified proofs.

QEDAI is a research prototype for converting natural-language mathematical statements into Lean 4–verified proofs. It combines language models with proof planning, Mathlib retrieval, parallel proof search, convergence matching, Pareto-based candidate selection, and verifier-guided repair to explore more reliable and transparent automated theorem proving.

  • Proof planning
  • Mathlib retrieval
  • Parallel proof search
  • Convergence matching
  • Pareto-based selection
  • Verifier-guided repair

Hardware verification

QEDSoft

RTL, golden models and requirements in one Lean 4 semantic model.

QEDSoft is a research prototype that formalizes supported SystemVerilog RTL, MATLAB golden models, and natural-language requirements into a shared Lean 4 semantic model. It represents hardware behavior as explicit fixed-width transition semantics, supporting kernel-checked RTL-to-model equivalence and specification-property analysis. Alongside generated SVA checkers and bind files for standard simulation and formal-verification flows, QEDSoft provides auditable evidence and fail-closed diagnostics whenever a complete formal interpretation cannot be established.

  • SystemVerilog RTL
  • MATLAB golden models
  • Natural-language requirements
  • RTL-to-model equivalence
  • SVA checkers and bind files
  • Fail-closed diagnostics

Quantum error correction

Quantum_QED

Proof-carrying quantum recovery.

Quantum_QED is a research prototype for assessing how reliably quantum error-correcting codes and decoders handle noise. Its Lean 4 verification layer checks certificates that bound decoder failure probabilities under clearly specified noise assumptions, making the guarantees and their limits easier to audit. Simulation and published hardware data provide additional evidence for evaluating the approach.

  • Error-correcting codes
  • Decoders
  • Failure-probability certificates
  • Explicit noise assumptions
  • Simulation
  • Published hardware data

Portfolio risk

AxiomVest

Portfolio proposals checked against policy rules in Lean 4.

AxiomVest is a research prototype that combines market data, statistical models, and portfolio optimization to help brokers monitor client portfolios and identify risks. Lean 4 formally verifies allocation proposals against defined policy rules before broker review, making each proposal traceable to its inputs and constraints.

  • Market data
  • Statistical models
  • Portfolio optimization
  • Policy rules in Lean 4
  • Broker review