Computer Science Labs · Open Exploration
Systems Research &
Computational Laboratories.
Fundamental computer science investigations bridging distributed state replication, formal verification, and hardware-efficient neural execution engines. All software is released under permissive open-source licenses.
4Active Labs
100%Open Source / Permissive
[CS-SYS-01]Distributed Systems & Runtimes
[ACTIVE]Orbit-VM: High-Concurrency Deterministic Execution Engine
LEAD: Matrika RegmiACTIVE: 2024 - PresentDOI: 10.5281/zenodo.georbit.91024
Microsecond-latency deterministic execution sandbox for parallel state transitions and AI agents.
FORMAL BASIS:
Petri-net concurrency verification, Lamport vector timestamps, software transactional memory (STM) invariants.
Key Performance Verifications:
- Zero divergence observed across 10^10 simulated adversarial network partitions.
- Achieved 140,000 transactions per second per execution shard with linear multi-core scaling.
- Published as 100% open-source runtime under Apache 2.0 / MIT licenses.
[CS-SYS-02]Distributed Consensus
[ACTIVE]Aether: Asynchronous Leaderless DAG Consensus Protocol
LEAD: Ayaan SaifiACTIVE: 2024 - 2025DOI: 10.5281/zenodo.georbit.82019
Leaderless directed acyclic graph consensus providing asynchronous safety and sub-second finality.
FORMAL BASIS:
Partial-order causal history graphs, Byzantine Quorum systems, probabilistic liveness bounds.
Key Performance Verifications:
- Demonstrated 240ms median finality over global multi-region WAN links.
- Resilient to arbitrary network delays with formal proof of safety under asynchronous conditions.
- Adopted as the foundational consensus layer for Orbit's enterprise telemetry platforms.
[CS-AI-03]Autonomous Intelligence
[ACTIVE]Eigen-Kernel: Sparse Tensor Compiler for Neural Reasoning Systems
LEAD: Matrika RegmiACTIVE: 2024 - PresentDOI: 10.5281/zenodo.georbit.71402
Optimizing code generator translating dynamic reasoning graphs into hardware-tailored sparse kernels.
FORMAL BASIS:
Polyhedral loop compilation, sparse matrix CSR/ELL format optimizations, tensor contraction algebra.
Key Performance Verifications:
- Achieved 3.4x higher token generation speed on sparse mixture-of-experts architectures.
- Cut VRAM footprint by 44% during multi-agent context retention benchmarks.
- Full PyTorch and ONNX JIT compilation backends open-sourced.
[CS-SEC-04]Applied Cryptography & Privacy
[ACTIVE]Poly-Proof: Sub-Second Recursive Polynomial Verifier
LEAD: Ayaan SaifiACTIVE: 2024 - PresentDOI: 10.5281/zenodo.georbit.61902
A folding-scheme zero-knowledge verification engine that validates complex off-chain cloud compute runs in sub-millisecond execution cycles.
FORMAL BASIS:
Nova folding schemes, multivariate sum-check protocols, elliptic curve cycle pairings.
Key Performance Verifications:
- Sub-40 millisecond proof generation for 65,536 R1CS constraints on consumer silicon.
- 48-byte constant-size recursive proof verified in 1.8 milliseconds on standard runtimes.
- Zero trusted setup ceremony required; publicly auditable mathematical guarantees.