Resources
Learn it, then verify it yourself.
Two halves. On the left, how to build with the C++. On the right, the proof: the numbers we recover and the real public files anyone can re-download and re-hash.
Help & documentation
Documentation and tutorials
From tarball to running code in three steps.
Every delivery is header-only C++17 with a portable build script and a self-checking demo. There is nothing to install beyond a compiler and Eigen.
- 1. Compile the demo with ./compile-native.sh, tuned to your CPU.
- 2. Run it. It recovers a known answer, so you confirm the math on your own machine.
- 3. Include the module header in your own code and call it directly.
# 1. compile a demo, native to your CPU
$ ./compile-native.sh genomics_demo
# 2. run it, it recovers a known answer
$ ./genomics_demo
[PASS] GC content 0.3797
[PASS] Ts/Tv chrMT 22.91
# 3. use it in your own code
#include "genomics/djehuti_popstruct.hpp"
Methodology and proof
We prove it twice: planted truth, then real public data.
First, every module recovers a value we planted, so the math is checked, not just run (50 / 50 checks pass, built NaN-honest so a silent bug fails its assert). Then we run the same engine on real public files a domain expert recognizes, and pin every input by SHA-256 and source URL in a tamper-evident log. You can re-download the files and re-hash them.
| Domain | Real data (source) | Engine reproduces |
|---|---|---|
| Genomics | 1000 Genomes chrMT / chrY, super-pop panel (EBI) | GC 37.97%, Ts/Tv 22.91 (== bcftools), PC1 splits African / non-African, Fst AFR/EAS 0.088 |
| Climate | NASA GISTEMP, NOAA Mauna Loa CO₂ | warming +0.081 °C / decade, anthropogenic signal detected (CO₂ fingerprint) |
| Quant | FRED S&P 500, Treasuries, WTI | 99% one-day VaR 3.40% (empirical 3.43%), 2y / 10y cointegrated, diversification frontier |
| Fraud (AI) | ULB credit-card, 284,807 real transactions (OpenML) | unsupervised ROC-AUC 0.95, recall at 1% budget 60%, no labels |
# downloads real public files, runs the engine,
# writes a SHA-256 pinned evidence log per domain
$ bash examples/realdata_check.sh
genomics ALL-GREEN
climate ALL-GREEN
quant ALL-GREEN
fraud ALL-GREEN
# verify any input the same way we did
$ sha256sum cov2.fasta
Public sources
- NCBI, SARS-CoV-2 reference genome
- 1000 Genomes Project (EBI)
- NASA GISTEMP surface temperature
- NOAA Mauna Loa CO₂
- FRED, Federal Reserve economic data
- OpenML, credit-card fraud (ULB)
The evidence log lists every file with its SHA-256 and source URL, and a manifest hashes each log. The logs cannot be quietly edited.
Case studies and use cases
One module, many industries.
A foundation is rarely single-purpose. These are realistic ways teams build on each module. They are illustrative, the decisions and the product remain yours.
Climate, beyond climate
Agriculture and water
The trends, extremes, and attribution engine drives crop-season planning and yield forecasting, and flood and drought return levels for utilities. Same code, different field.
Quant, beyond finance
Any portfolio of risks
Extreme-value risk and correlated scenarios apply to insurance, energy trading, and supply-chain exposure, not only trading desks.
Genomics
Research and breeding
Pop-structure PCA and Fst scans serve human-cohort research as well as crop and livestock breeding, on the same packed genotype store.
Neural / AI, especially
Anomaly across sectors
The autoencoder and isolation-forest anomaly stack that scores ROC-AUC 0.95 on real card fraud is the same one that flags faults, intrusions, and quality defects. See the AI suite
FAQ
Questions, answered.
Do I get the actual source code?
Yes. You receive the licensed module's source, header-only C++17, watermarked to you. You compile it yourself.
Can I ship it inside my own product?
Yes. The license grants OEM and embed rights: compile it into your product and distribute that product to your customers, including as SaaS. You may not resell the source as a competing SDK.
What are the dependencies?
Eigen only on the baseline build. Optional formats (NetCDF, zlib for gzip) sit behind explicit flags. No Python, no hidden runtime.
What hardware does it need?
Any modern x86-64 CPU with AVX2. AVX-512 is used automatically where present. The build auto-tunes cache-block sizes to your machine.
How do I trust the math?
Every module recovers a known answer (50 / 50 checks), and we validate on real public data with SHA-256-pinned logs you can reproduce. See the methodology above.
What support is included?
Thirty days of email questions from delivery. Ongoing support, modular updates, new add-ons, and performance tuning are available on a separate agreement.
How do I pay, and when do I get the code?
Wise, bank transfer, or crypto. We show the SHA-256 of the exact build during the demo and deliver the source once payment confirms. Crypto confirms in minutes.
Want to see it run on your data?
We will run the demo and the recovery checks live, and you can re-hash every public input yourself.