“I swapped three small NumPy helper scripts for Aion maths utilities in our lab notebook. Same outputs, fewer lines, and I could actually explain every step in the paper appendix.”
Mariam K.
Graduate researcher · applied ML
Open tools for the research stack
Aqwel AI turns opaque ML workflows into auditable, reproducible pipelines—with open libraries, datasets, and documentation built for researchers and engineers.
pipeline.py· PyTekt v0.2.0 Pipeline
223.4M rows/sApple Silicon Metal
1import pytekt as pt2from pytekt.datasets import load_benchmark3from pytekt.former import FormerBlock, FastAttention4from pytekt.metrics import evaluate_run, accuracy56# 1. Reproducible seed & hardware acceleration7pt.seed(42)8device = pt.hardware.detect_accelerator() # Metal / SIMD910# 2. Ingest 1,000,000 tabular research records11dataset = load_benchmark("nyc-taxi-2.96m", scale=1_000_000)12X, y = dataset.to_cache_aligned_tensors(align=64)1314# 3. Compile AST-to-SIMD transformation kernel15@pt.jit(target="simd", vectorize=True)16def preprocess(row):17 return (row.fare * 2.0 + row.dist) / (row.tip + 1e-5)1819features = preprocess(X)2021# 4. Neural optimization & metric convergence22model = FormerBlock(d_model=128, n_heads=8, device=device)23loss, preds = model.fit(features, y, epochs=50, lr=1e-3)2425# 5. Auditable telemetry26metrics = evaluate_run(y, preds, track_memory=True)27print(f"acc: {metrics.accuracy:.1%} | loss: {loss:.4f} | rate: 223.4M/s")
Telemetry & MetricsExit code: 0
Accuracy99.4%+2.4% vs base
Loss0.0142Converged
Speedup450.3×AST-to-SIMD
Accuracy CurveLoss Decay
Epoch 1Epoch 25Epoch 50 (99.4%)
✓ 1,000,000 records aligned to 64-byte cache
✓ AST-to-SIMD JIT kernel compiled in 0.38ms
→ Finished in 4.48ms · throughput: 223.4M/s
500+
Primitives
Maths, physics & ML
38
Pre-indexed Datasets
Up to 1M rows/set
100%
Deterministic
Zero stochastic drift
Apache 2.0
Open Core
Inspectable license
v0.2.0
Active Release
PyTekt stream
Introducing OpenKL
A new open library from Aqwel AI. Built for transparent, reproducible research workflows. We're putting the finishing touches on it.
OpenKL
pip install openkl
Aion is now PyTekt
Our open-source scientific computing library is entering a new chapter. A refined identity tailored for Python-native workflows, expanded scientific toolkits, and verifiable intelligence.
$pip install pytekt
Read documentationLEGACY NAMESPACEv0.2.0
Aion
Package: aqwel-aion
Initial foundation for reproducible research, NumPy Former, and deterministic matrix algebra.
Fully backward-compatible alias maintained.
ACTIVE RELEASEv0.2.0
PyTekt
Package: pytekt
Next-generation architecture: 500+ primitives, physics & astronomy toolkits, native C++ dispatch, and CLI workflows.
Recommended package for all new installations.
quickstart.py
# Install: pip install pytekt
import pytekt as pt
# 1. Deterministic maths & tensor kernels
A = pt.maths.tensor([[1.0, 2.0], [3.0, 4.0]])
det = pt.maths.determinant(A)
# 2. Scientific & physics simulation
pendulum = pt.physics.Pendulum(length=1.0, mass=0.5)
orbit = pt.universe.compute_orbit(semi_major_axis=1.496e11, eccentricity=0.0167)
# 3. Composable Transformer former
former = pt.former.TransformerEncoder(embed_dim=512, num_heads=8)
print(f"PyTekt initialized • Determinant: {det:.2f}")Pythonic Foundation
Engineered directly for Python without opaque wrappers. Composable NumPy, SciPy, and PyTorch interop across 500+ primitives.
Architectural Rigor
Modular architecture with neural blocks, deterministic physics integrators, and auditable pipelines.
Backward Compatibility
Existing Aion workflows continue working seamlessly with drop-in aliases and zero breaking changes.
Open & Verifiable Core
Every algorithm and benchmark is fully open-source, deterministic, and inspectable under the Apache 2.0 license.
Engineered for clarity
Stateless operators, auditable I/O, and over 500 primitives engineered directly for Python.
MODULE 01
Mathematics & Stats
71+ deterministic maths primitives for linear algebra, covariance, and ML kernels.
DETERMINISTIC • VERIFIABLE
MODULE 02
Physics & Astronomy
Simulations, celestial coordinates, and orbital mechanics with optional C++ dispatch.
DETERMINISTIC • VERIFIABLE
MODULE 03
Research Stack
Embeddings, RAG primitives, providers, and a NumPy Former in one Python surface.
DETERMINISTIC • VERIFIABLE
Version comparison
From focused ML scripts to a full-spectrum scientific computing platform.
| Capability | v0.1.9 | v0.2.0 (PyTekt) |
|---|---|---|
| Focus | Providers, tools, RAG, Former, native fast_* | + physics, universe, vision, core ML, cache, serve/db, Hub |
| Modules | ~15 documented core groups | 50+ modules / 500+ primitives |
| Science | — | physics, universe |
| Vision / ML | Visualization + evaluate | vision, preprocessing, models, metrics, hyperopt |
| Platform | CLI, config, I/O | + serve, db, store, tracker, usage, hub, doctor |
| Upgrade | pip install aqwel-aion | pip install -U "aqwel-aion>=0.2.0" |
Simple, expressive syntax
Consistent APIs for physics, astronomy, computer vision, and machine learning.
Natural-language physics query
Solve classical physics tasks with aion.physics — or run a pendulum simulation.
python
from aion.physics import solve_physics_query, simulate_pendulum
result = solve_physics_query("kinetic energy mass=2 velocity=3")
print(result.output_value) # 9.0
sim = simulate_pendulum(1.0, 0.2, steps=100)
print(sim.summary["small_angle_period_s"])Command-line workflows
Health checks, local experimentation hubs, and native acceleration diagnostics built into the CLI.
Install
Install PyTekt v0.2.0
Pull the latest PyTekt release and optional extras for vision, serve, or the full stack.
$ pip install -U "pytekt>=0.2.0"
Extras: [vision] [serve] [db] [full]
CLI
Environment health check
Verify Python, numpy, optional deps, and native backends before experiments.
$ pytekt doctor
Also: pytekt welcome · pytekt info
Hub
Local Hub dashboard
Browse modules, dependency status, and run snippets in the browser.
$ pytekt start
Opens http://127.0.0.1:3000
Science
Physics & universe CLIs
Run science toolkits from the terminal without writing a notebook first.
$ pytekt physics --help && pytekt universe --help
Also: pytekt vision · pytekt usage
Built for researchers
Designed for independent researchers, labs, and modern development teams.
Scientific research
Reproducible experiments with verifiable mathematics and transparent seeds.
Open datasets
Download structured CSV & JSON datasets without auth for benchmarking.
Engineering solutions
Specialized bot engineering, automation, and custom model integrations.
Datasets
Explore & download open data
88 datasets across 8 domains. Up to 1M rows per dataset. Free for research in CSV and JSON formats.
Time seriesTabularNLPFinanceSensorHealthcareRetailGeospatial
Wide structured tables tailored for ML experiments, baseline benchmarks, and verifiable research pipelines. Direct instant download without account requirements.
Ecosystem
Stories and prompt engineering
Fresh announcements on our Blog, paired with Lattice—a curated directory of battle-tested prompts.
Open & production ready
Built for teams running reproducible AI research with reliability, transparency, and control at every layer.
Mission
Building for the research community
We ship open infrastructure so scientific discovery remains inspectable, reproducible, and universally accessible.
PILLAR 01
Open tooling
Stateless maths, explicit I/O, and auditable primitives—libraries you can cite in a paper without vendor lock-in.
AUDITABLE • REPRODUCIBLE
PILLAR 02
Transparency first
Clear computation paths, documented optional accelerators, and APIs designed for inspection—not black-box convenience.
AUDITABLE • REPRODUCIBLE
PILLAR 03
Reproducible workflows
From seeds and checksums to layered configs—research steps that stay traceable from raw data to published claims.
AUDITABLE • REPRODUCIBLE
Research
Technical notes from the lab
Mechanics-first articles on machine learning, computer vision, and reproducible experimentation.
Collaborators
Partners & sponsors
Collaborations that help us deliver open research, education, and AI-native software.
Community
Voices from the community
Perspectives from researchers, engineers, and educators applying our open libraries and datasets.
Aion
“The evaluate module saved us a weekend—accuracy, F1, and ROC-AUC in one place without wiring sklearn boilerplate for every ablation run.”
Jonas R.
ML engineer · startup
“Wide synthetic tables with predictable seeds are perfect for teaching feature engineering. Students download CSV, preview rows in-browser, and reproduce the same splits every time.”
Elena V.
Instructor · data science
“The Aion docs site is what PyPI pages wish they were—searchable functions, copy-paste examples, and module boundaries that match the import paths.”
Dev P.
Backend engineer · open source
“The CNN matrices article finally connected convolutions to linear algebra the way I needed for a internal tech talk. No hand-wavy diagrams—just mechanics.”
Alex T.
Computer vision engineer
“Core install is lightweight; we added [full] only on the training machine. That layered dependency story is rare and honest.”
Sofia N.
Research software engineer
FAQ
Frequently asked questions
Everything you need to know about our open libraries, licenses, and architecture.