We build it. You own it.
Custom software and private AI, built for how your business actually works and running on your own infrastructure. You own the source and the weights outright. Awareness Software Group.
Analytics platforms, legacy replacement, integrations, and AI models trained on your data. Twenty years of engineering, two US patents, and a decade building for government, military and intelligence customers.
Fixed-price steps, small increments, and no lock-in. Keep us on to run and extend it, or take the source and the documentation and run it with your own team. Both are fine with us.
Contract engineering · with or without AI
Software We Build
Plenty of what we are asked for has no model in it at all: a platform, an integration, a pipeline, a system that has to be right. We take that work, and we take it end to end. You own what we build, source and all.
Data & Analytics Platforms
Custom analytics built for how your business actually works: ingestion from the systems you already run, a warehouse or lakehouse that fits your scale, modelled metrics people trust, and dashboards or reporting your team uses without asking us. Built to be handed over, not hosted hostage.
Legacy Modernization & Replacement
Systems that still run the business and are holding it back: an unsupported platform, a vendor gone end-of-life, a database nobody dares touch, or an Access file that quietly became infrastructure. We update in place where that is the cheaper answer and replace where it is not, incrementally, with the old system running the whole time. Big-bang rewrites are how these projects fail, and we do not start one.
Integration & Automation
Getting your systems to talk to each other: APIs, ETL and event pipelines, third-party services, legacy databases, internal tools, and the unglamorous glue in between. Including the cases where the vendor says it cannot be done.
AI Platform Implementation
You have picked an AI platform, or you want AI inside a product you already have. We do the implementation and the integration: retrieval over your documents, assistants and agents wired to your real systems, evaluation so you know whether it works, and guardrails so it fails safely.
Custom Applications
Internal tools, web applications, back-end services, data products, and the one-off system nobody sells off the shelf because it is specific to you. Delivered in small, working increments you can see and steer.
Assessment & Rescue
A second opinion on an architecture, a build that has stalled, or a vendor claim you want tested before you sign. We are comfortable telling you the honest answer is to do less, or nothing, and the public research on this site is a fair sample of how hard we test our own conclusions before we stand behind them.
All six in detail, with how we work and where it runs →
Any of these can be built inside the boundary you answer for: on-premises, in your own cloud, or fully air-gapped, under HIPAA, CJIS, GDPR or export control. Our founder spent more than a decade building systems for US government, military, and intelligence customers, where data sovereignty and reliability were the requirement rather than a feature.
The proof we would point at: the AI work on this page runs on a platform we designed and built ourselves, end to end, from data ingestion through evaluation to private deployment. It is a real system carrying real client data, and it is the same engineering you would be buying. Fixed-price steps, small increments, and you own the source.
Measured, not asserted
We don't say it's better. We measure it.
Whatever we build, we would rather show you a number than a demo. Here is ours on a public benchmark of 510 real lawyer-annotated commercial contracts: a base model against the same model fine-tuned, scored on 99 contracts it never saw in training. 48 measured runs across six open-weight models, every one improved.
62%
Off-the-shelf model
85%
Fine-tuned on the data
48
Measured runs, six models. Every one improved
And the part most vendors won't show you: along the way the same pipeline produced a 72-point gain, a 45-point gain and an 88-point gain that did not survive checking. In two of them the baseline model had never actually answered, so it was being scored on a difference in formatting rather than on capability. In the third, the prompt never named the field being scored, so what got measured was the model learning our file format. Our checks caught all three before they left the lab, and we published exactly how so you can run the same checks on anyone else's numbers. A results table where nothing was ever rejected is not a measurement, it is a brochure.
Measured on short-field extraction (document name, agreement date, effective date). Long clause spans need a different metric and we don't claim a number for them. These are public contracts, not yours. The assessment measures your data. Results derive from the Contract Understanding Atticus Dataset (CUAD) v1, The Atticus Project, CC BY 4.0.
When the problem needs a model
Sometimes the answer is not code. Own that too.
Some problems are not a program anyone can write: reading messy documents, answering from twenty years of institutional knowledge, judging something an expert judges by eye. Those need a model trained on your data rather than software written against a spec. We build those the same way we build everything else, and you own the result outright: no per-use fees, no data leaving the building, no lock-in, and no AI team required on your side.
You Own the Weights
The model, the training pipeline, and the evaluation harness are yours at the end of the engagement, not licensed and not rented. It keeps working whether or not we do.
Your Data Never Leaves
Training and serving happen on infrastructure you control: on-prem, your own cloud, or fully air-gapped. Nothing is sent to a third-party API, with deterministic offline redaction and a tamper-evident audit log on every engagement.
Your Data Becomes an Asset
We ingest your real, messy knowledge (files, databases, tickets, code, chats, PDFs, logs) and structure, label, timestamp and version it into a clean dataset you own forever. It outlives any single model.
Proven, or You Don't Pay
Backed by our Lift Guarantee: your model beats your baseline on your own tasks, measured in a written report, or you don't pay for the model. Evidence, not claims.
Start free with a data-readiness scorecard you run yourself · then fixed-price steps from $5k, from your first owned model to a managed, growing AI layer.
New: private design-partner preview
Awareness Verify. Break your AI before you ship it.
Ordinary software tests can pass while an AI system changes how it uses tools, handles approval, follows poisoned context, responds under pressure, or behaves after a model or prompt update. Awareness Verify continuously and adversarially tests AI agents and models, compares their behavior with known-good releases, and can stop a serious regression before it reaches users.
Adversarial Testing
Replay security, behavioral, tool, MCP, authorization, and reliability scenarios against the system that is actually being shipped.
Regression Evidence
Compare a candidate with an approved baseline or a previous run, and see which findings were introduced, which persist, and which were resolved.
Release Gating
Turn reviewed policy into a machine exit code, so a serious new failure can stop a deployment while preserving the evidence needed to investigate it.
A 30-Day Pilot
One authorized staging AI system, founder-led onboarding, and no software fee for the initial pilot. The first cohort is limited to 10 organizations.
In a synthetic local demo, a deliberately vulnerable target is blocked with 146 findings across 70 scenarios; the remediated target passes with 0. No third-party model, provider, or production system is contacted.
Who We Are
Deep technical experience and mission-focused execution.
Founder-led by an engineer from high-assurance environments
Awareness Software Group was founded by Joshua Bolick, who has spent 20+ years as an engineer, including more than a decade at a satellite communications company building systems for US government, military, and intelligence customers, and earning two US patents as a named inventor. In those environments, data sovereignty, security, and reliability aren't features. They're the requirement. That conviction is the reason ASG builds private AI you own and run on your own infrastructure.
Don't take our word for it: check the background yourself on LinkedIn.
Our own proprietary platform, so delivery is engineered, not artisanal
We built our platform from scratch: an end-to-end pipeline that carries any-shape business data through profiling, dataset-building, fine-tuning, evaluation, governance, and private deployment. It runs on trusted open-weight models, which is exactly what keeps your model owned and lock-in-free, but the engine is ours. That's what lets us turn your knowledge into a proven, owned model in small, low-surprise steps, without you needing an AI team of your own, with capability that compounds on every engagement.
Also from us
How we know which numbers to trust: our open research
We run a public research program into training AI models more cheaply and efficiently, using models small enough that anyone can reproduce every number on a laptop. We publish all of it, including the approaches that turned out not to work, because those are the ones that narrow the search. It exists because measurement is the hard part: knowing which numbers are safe to rely on is most of the job. That is the same discipline we bring to a client's numbers. Every study → · the plain-language versions →
Also from us: immutable threat-intelligence snapshots
A separate, smaller line of work: time-stamped, auditable dataset snapshots built from authoritative CISA sources, designed so an ML team can reproduce a result months later and tell data drift apart from model regression. The same instinct behind our AI work: if you can't reproduce it, you can't claim it. Snapshots, formats & pricing →
