For Alaska oil, gas, mining & resource operations
Resource operations in Alaska generate regulatory, safety, and procurement documentation at a scale no team can read. We build agents that search it, structure it, and cite it, so the answer takes seconds and can be defended.
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
It makes an unreadable document estate searchable and it structures the reports coming in from the field. Permits, plans of operation, environmental commitments, regulatory correspondence, and contractor submittals easily run into tens of thousands of pages across agencies. An agent answers a plain-language question about a commitment or a condition and cites the exact page it came from, which is the only form of answer that is useful in a regulatory conversation. On the field side, incident reports, inspection findings, and shift handovers arrive as free text and photos, and an agent turns them into structured records that can be trended. Neither replaces a compliance manager or an HSE lead. Both give them back the hours they currently spend searching.
The operating reality
North Slope, Cook Inlet, and interior mining operations share a profile: multi-agency oversight, long procurement lead times, rotating crews, and a cost of error measured in days of production.
Commitments live across state and federal permits, plans, and correspondence in different formats from different eras. No single person has read all of it and no index covers all of it.
A part that is a two-day order in Texas may be a season away here. Expediting is a real discipline and it runs on document chasing.
Two-week rotations mean the person who saw the problem is off shift when it matters. Shift handover quality is an operational risk, not a formality.
A planned turnaround has a fixed window and a readiness checklist thousands of lines long. Anything that surfaces an incomplete item earlier is worth real money.
Six workflows we build
Weighted toward the document work that currently blocks decisions.
01
02
03
04
05
06
First 30 days
When a superintendent has to wait two days to learn whether a condition allows something, that delay repeats every week and nobody logs it as a cost.
We inventory the permit, plan, and correspondence estate, including the scanned material from before anything was digital, and report honestly on what is and is not machine-readable.
Every answer carries a citation to document and page. An answer without a source is not an answer in a regulatory context, so the system refuses to give one.
We run it against questions your compliance team already knows the answers to and report accuracy, including the questions it gets wrong.
What you own at day 30
A cited search agent over your document estate running in infrastructure you control, an accuracy report against a known question set, the source code and index configuration, and a documented process for adding new permits as they are issued.
What we built, in the open
A permit question resolved from memory is worthless the moment an agency asks where it came from. The hard part of building a searchable permit estate is not retrieval, it is refusing to answer when the underlying record does not actually say what the answer claims.
Long-context research, which is the reason a document estate measured in tens of thousands of pages can be reasoned over at all. The benchmark harness is published alongside the numbers.
A gate that requires a generated answer to be grounded in a retrieved passage and to decline when it is not. We publish its false-accept rate, which is the figure that matters when the answer goes to a regulator.
Our own memory server, including an offline deployment. For an operator who cannot send permit and incident data to a third-party service, that is the difference between a project and a non-starter.
Public, with the harness. It is why we quote an error rate against your own question set instead of describing the system as accurate.
Scope, stated up front
In a regulated operation, a tool that guesses is a liability. Three places where this one stops and a person starts.
It is not a compliance determination
It finds and cites what a document says. Whether you are in compliance is a judgment your compliance manager and counsel make, and we will not blur that boundary to make a better sales page.
Bad scans stay bad
A 1990s fax of a fax has a real error rate no system fixes. We report which parts of your estate are reliably readable and which are not, before you rely on it.
It does not touch process safety systems
No control system integration, no safety instrumented function, nothing in the path of a shutdown. Document and reporting layer only.
Where your data goes
Permit estates and incident records are regulatory artifacts. Five commitments we put in the agreement.
Yes, and for most operators it should. We build offline memory infrastructure as part of our own research, so running the index and the agent inside your tenant or on your own hardware is a configuration choice rather than a custom engineering project.
That is the normal case. We report per-document what is reliably machine-readable and what is not, rather than pretending an unreadable scan produced a trustworthy answer. Where extraction quality is poor, the system says so instead of guessing.
Because it comes with a citation to the document and page, and checking it takes ten seconds. We also test against a set of questions your compliance team already knows the answers to, and we publish the error rate from that test to you before go-live. A system that cannot tell you when it is unsure is not usable in a regulated environment.
Yes, through their APIs where available. The document layer usually sits alongside rather than inside the ERP, because the ERP was never designed to hold a permit estate and forcing it to is more expensive than the problem.
Prequalification packages, HSE reporting, and turnaround documentation are the same problems from the other side of the contract. Service companies often see faster payback because their documentation burden is imposed by multiple different clients with different formats.
It tracks commitments and reporting obligations from the permits and flags upcoming deadlines with the source condition attached. Your regulatory affairs team still owns every submission. The value is that nothing gets missed because it was buried on page four thousand of a document nobody had time to reread.
We will show you what a cited search over your own documents returns, and where it fails. Sixty minutes, no cost.
Anchorage, Alaska · info@northtek.io · (907) 903-4353