The infrastructure of discovery.
We focus on high-performance mathematical tooling, accelerated columnar DataFrame engines, and open libraries for precise experimentation and verifiable research.
PyTekt Library
PyTekt is an open-source Python scientific computing and AI research library by Aqwel AI. Designed as the official evolution of Aion, PyTekt delivers 500+ composable primitives across deterministic tensor operations, physics and celestial mechanics, transformer modules, and auditable experimentation pipelines.
Key Capabilities
- 500+ stateless mathematical operators & deterministic tensor kernels
- Classical mechanics, celestial orbits, and physics integrators
- Composable Transformer Former neural architectures
- Multi-provider LLM tooling, RAG pipelines, and agent loops
- Local experiment tracking, telemetry dashboards, and system diagnostics
- Full drop-in backward compatibility with legacy Aion workflows
OpenKL Engine
OpenKL (openkl.DataFrame) is a high-throughput, hardware-accelerated tabular data processing engine written in C++20 for Python. Engineered to eliminate traditional DataFrame bottlenecks, OpenKL features 64-byte cache alignment, AST expression fusion, SIMD vectorization (AVX-512 & ARM NEON), and multi-GPU streaming dispatch—delivering up to 573× speedups over standard Python DataFrames.
Key Capabilities
- 64-byte cache-aligned columnar memory buffers for AVX-512 & ARM NEON
- JIT expression compilation via AST fusion, eliminating temporary intermediate allocations
- Multi-GPU & multi-stream asynchronous kernel dispatch for large tabular workloads
- Seamless Pandas API interoperability with zero-copy Arrow memory buffers
- Parallel chunked CSV/Parquet parsers with auto memory-mapped IO
- Fast predicate pushdown, grouped aggregations, and rolling transforms