For Alaska federal contractors, 8(a) firms & defense suppliers
Federal contracting rewards proposal throughput and punishes compliance mistakes. We build agents that triage solicitations, generate compliance matrices, and draft from your real past performance - and we scope the controlled unclassified information boundary before writing a line of code.
Built in Anchorage · we publish our source code · no long-term lock-in
Written by Kristian Baer, Northtek · Anchorage, Alaska · Updated 2026-08-21
The short answer
On most of it, yes, provided you separate the data classes first. Publicly posted solicitations, your own marketing material, and your internal past performance records are generally unrestricted and are where nearly all of the proposal time is spent. Controlled unclassified information is different: it carries handling requirements under your contract clauses and most commercial AI services are not authorized destinations for it. The practical answer is an architecture that keeps CUI out of any unapproved system entirely, while agents work freely on the unrestricted material. A vendor who does not raise this distinction before selling you a tool is either unfamiliar with your obligations or hoping you are. We map the boundary in writing during scoping, and if part of your workflow cannot be automated compliantly we say so rather than building it anyway.
The operating reality
Alaska federal contractors compete nationally from a small labor market, often across JBER, Eielson, Fort Wainwright, Coast Guard installations, and remote sites, with proposal teams too small for the pipeline they want.
A missed Section L instruction can eliminate a proposal before anyone reads the technical approach. The compliance matrix is not paperwork, it is the gate.
CPARS records, final reports, and closeout documents contain everything you need to write a strong proposal, scattered across drives nobody has indexed.
Go/no-go discipline is the difference between a healthy pipeline and a team burning out on bids they were never going to win. Most firms decide by instinct.
If your contracts carry CUI handling requirements, tool selection stops being an IT preference and becomes a contractual obligation with real consequences.
Six workflows we build
All of them operate on unrestricted data unless your environment is assessed and approved for more.
01
02
03
04
05
06
First 30 days
It is unrestricted data, it is high value, and it makes every subsequent proposal faster whether or not you automate anything else.
Before any development we classify your document estate: unrestricted, business sensitive, and CUI. That map determines the architecture and it goes in the scope document your contracts manager signs.
CPARS records, final reports, and closeout documents become a searchable, citable library. Every answer names the document it came from.
On your next real bid, the library runs alongside your normal process. Your capture manager judges whether it surfaced better evidence faster, and if it did not, we say so.
What you own at day 30
A cited past performance library in an environment matched to your data classification, a written data boundary document, the repository and configuration under your ownership, and a documented list of the workflows we determined should not be automated in your environment.
What we built, in the open
Every other vertical treats hallucination as an accuracy issue. On federal work it is a legal exposure, because a past performance claim your documents do not support is a false statement to the government. That changes what the system has to be built out of.
A verification gate that requires every factual claim in a draft to trace to a source document, and marks the gap rather than filling it when no source exists. We publish its false-accept rate, including what it misses.
A language built so an agent's steps are auditable. If a contracting officer or your own counsel asks how a statement got into a proposal, the path exists rather than being reconstructed after the question.
A prompt-injection scanner. Solicitation packages and teaming submissions are documents from outside your firm, and treating them as trusted input is how an agent gets steered.
All of it is public with its commit history. In a market where every firm asserts accuracy, a published error rate is the only version of that claim you can check.
Scope, stated up front
On federal work the wrong automation is worse than no automation. Three limits we build in deliberately.
It will not invent past performance
Every factual claim traces to a source document. Where evidence does not exist the draft says so and leaves the gap visible. A model that fills gaps confidently is a liability in this industry, not a feature.
It is not an authorized CUI environment
Unless your environment is separately assessed and approved, we build for unrestricted data and we keep CUI out of it by design. We will not tell you a commercial tool is fine because it would be easier to sell.
It does not replace your capture manager
Win themes, pricing strategy, and the decision to bid are human calls. We remove the document assembly around them.
Where your data goes
Data classification precedes development, and the boundary is enforced technically rather than by training. Five terms we put in the scope document your contracts manager signs.
Yes, on the unrestricted portion of your work, which is most of the proposal effort. Public solicitations, your capability statements, and your own past performance narratives usually sit outside CUI. We classify your estate during scoping and build only on what your contract clauses permit. If you need automation on the CUI side, that requires an assessed environment and it is a separate, larger conversation we will scope honestly rather than hand-wave.
Because every factual claim in a draft carries a citation to the source document, and claims that cannot be sourced are flagged rather than written. We also publish FACTGATE, our verification gate, including its false-accept rate. A vendor who tells you their system never hallucinates is telling you they have not measured it.
Yes, and small firms often benefit most, because the compliance and documentation burden is roughly fixed while the proposal team is not. A three-person capture team gets the biggest proportional gain from automated compliance matrices.
No, and we would not sell that. It drafts from sourced material against a compliance matrix, which is the mechanical portion. Win themes, discriminators, and the judgment about what an evaluator actually wants stay with your capture manager, and proposals written entirely by a model read exactly like proposals written entirely by a model.
They stay. We consume their feeds and add the reasoning layer - scoring against your actual capability and capacity rather than just alerting you that something posted.
It adds a data governance layer above the federal one. The corporation may have policies about where subsidiary data can reside, and those govern the architecture. See the Alaska Native corporations playbook for how we handle that, and we will design to both regimes rather than asking either one for an exception.
Above them, not instead of them. Platform partners run identity, licensing, integration, and uptime across Dynamics, Azure, and Microsoft 365. We build the agent layer that reads solicitations and past performance and writes into whatever they maintain. The two do not compete, and a firm that proposes replacing a working platform estate in order to add AI to it is describing a bigger project than the problem requires.
We will build the compliance matrix live and show you what a cited draft looks like. Sixty minutes, no cost, and we will name what we cannot compliantly touch.
Anchorage, Alaska · info@northtek.io · (907) 903-4353