Private AI for the work that matters
We build AI agents for the repetitive work buried in your reports, documents, and requests. They run inside your own systems, ask before they act, and keep a record you can audit.
BOUNDARY: INTACT. REGION: US / EU. DATA PLANE: CUSTOMER-OWNED. TRACES: SIGNED. AUTONOMY: EVIDENCE-GATED.
Summaries, checklists, and drafts start from the records already in your systems, not a blank page.
Missing records, overdue actions, cost drift, and exceptions get flagged before the monthly review or the audit.
Agents recommend first. They act only after the rules, the approval, and the record are clear.
01 · The problem
A single claim is spread across the policy system, intake forms, PDFs, adjuster notes, email threads, and the document store. They rarely line up when someone needs to decide. Teams spend hours assembling evidence and chasing missing documents instead of moving the file.
Policy data, forms, correspondence, and notes each tell part of the story. No one sees the whole file fast enough to act on it.
Missing documents, overdue follow-ups, cost drift, and exceptions often surface after payout timing or audit posture is already at risk.
Your most experienced people know what matters, but that judgement is hard to share across teams, sites, and a growing backlog.
Built for teams where a missed detail moves cost, payout timing, compliance, or audit posture
We start where evidence is fragmented and the backlog is slow. One workflow first, one Assurance Record, then the next step when the evidence supports it.
02 · The platform
Bring together the records the file already depends on: policy systems, intake forms, PDFs, the document store, adjuster notes, and correspondence. We connect them inside your own cloud, datacenter, or approved region through the Boundary Runtime.
Agents draft the things claims teams lose time on: claim summaries, missing-document checklists, evidence packets, exception flags, and routing recommendations. People review the work before it moves the file.
The Evidence Layer points every answer back to the records it used, who approved it, what it cost, and what still needs attention. Managers get an Assurance Record they can replay, not a black box.
Once the evidence is good enough, the agent can take one approved step: post a missing-document checklist, route an exception, or update a record. Higher-risk steps wait until the Autonomy Ladder says it is safe.
03 · Trust Boundary
Every claim file, document, prompt, note, and trace stays inside your environment. We fit the deployment to your privacy rules, access policies, and data-residency needs, then run the controls through access you approve.
Claim files, documents, traces, notes, and agent runs stay inside your environment. Only usage metadata ever leaves.
We handle setup, monitoring, and support through the access you grant us, never around it.
Each request checks who can access what, masks sensitive details when needed, holds spending limits, and records which model was used.
An agent can only open the files and records the assigned user is already cleared to access. Nothing more.
04 · Evidence Layer
This is not a developer dashboard. The Evidence Layer shows managers what the agent prepared, which sources it used, what each run cost, what still needs approval, and what the Autonomy Ladder says is safe to automate next.
Each agent has one clear job, an owner, a scope, and an evidence gate before action
05 · Autonomy Ladder
Every workflow climbs the same Autonomy Ladder, where an agent earns more responsibility only as the Evidence Layer proves it out.
Answers and drafts. A human does everything.
The agent proposes; a human approves every action.
The agent recommends the next step to take.
The agent acts after a human signs off.
The agent acts within set guardrails; humans audit.
Scheduled and self-running; humans handle exceptions.
The governance horizon
Banned practices in force (from 2 Feb 2025)
Agency use & acquisition (OMB M-25-21 / M-25-22)
General application (from 2 Aug 2026)
WE BUILD FOR THISColorado, California, Texas obligations
Annex I product-integrated AI (from 2 Aug 2027)
06 · How we work
We start by understanding the workflow, not pitching a solution. Our engineers sit with the people who run claims, learn where the file slows down, and map the records your team already trusts. We come to learn the work in order to build it, not to write a report.
OUTPUT A clear, shared definition of the workflow, the evidence needed, and where an agent would actually help.
Together we pick one assurance-critical workflow first and agree what the pilot will prove: what success looks like, which sources count, who approves action, and where the agent must stop. We run it Shadow-First inside your environment: the agent observes and recommends before it is allowed to act, then takes one approved low-risk production step.
OUTPUT One workflow proven on real claim files, on a scope you signed off, with an Assurance Record.
With the Assurance Record in hand, we decide together how to continue: move up the Autonomy Ladder, expand to the next workflow, adjust the approach, or stop. The evidence makes the call, not a contract. If it is working, we plan the next workflows with you and build them out, one at a time.
OUTPUT A decision backed by the Assurance Record, on your terms.
Engagements
Hands-on service now, a product later. We develop and ship a solution built around your team, your systems, your privacy rules, and the approvals your operation needs. Each engagement is a step on the same path: land one workflow, prove it with an Assurance Record, and build the Evidence Layer every future workflow runs on.
A claims workflow depends on fragmented evidence and too much manual chasing, but agent runs cannot leak into outside AI tools.
In 30–45 days: one private claims workflow run Shadow-First inside your environment, where the agent prepares the work, shows the sources, and takes one approved low-risk production step. You get an Assurance Record: cost per case, approvals, failures, replay examples, and an autonomy recommendation.
A workflow is live, but processes change, prompts drift, and no one owns keeping the agents reliable and defensible.
Ongoing evals, replay tests, prompt and policy updates, incident review, and a quarterly read of the Evidence Layer on what to promote up the Autonomy Ladder next.
One workflow works; expanding across more claim workflows without losing control is the hard part.
A forward-deployed pod that adds new agents and connections inside your team, giving each workflow more responsibility only as the Evidence Layer supports it.
07 · The arc
Proqtor is early, and focused on purpose. Here is the plan, in plain sight.
One claims workflow, deployed inside your own environment and proven Shadow-First on real files before it touches production. We embed in your team and build it end to end.
One workflow at a time, each its own product. We go where an agent can affect cost, payout timing, compliance, customer commitments, or audit posture, and only when the last one has earned it on the Autonomy Ladder. For example: intake and triage, missing-document checks, evidence assembly, exception routing, and policy lookups. Adjacent regulated workflows stay open for pivot.
Under every workflow is the same Evidence Layer: cost per completed task, approvals, a replayable record of every run, and what is safe to automate next. That layer is the part we are building into a product. Hands-on service first, because that is how you learn a workflow honestly. A product second, because that is what the Evidence Layer is for.
This is the plan, not a finished product. Today, the work is hands-on, one workflow at a time.
ALBERT GARCÍA HERNÁNDEZ
FOUNDER
08 · About
Proqtor is built and run by its founder, Albert García Hernández, and the team works by one rule: the people who build your system work inside it. We are forward-deployed.
Forward-deployed is the model Palantir made its name on. The engineers who build the software work alongside your team, on your real claim files, and stay until it runs in production, instead of selling a license and walking away. We hold to that standard at founder scale: whoever plans your deployment also builds the agents, sets up the Evidence Layer, and signs off on the security setup. No junior bench, no plan handed to someone who has never seen your work.
And we come to where the work is. We serve teams across the US, Europe, and beyond, deployed in your cloud, your datacenter, or on-site, always inside the Trust Boundary that holds your data, and built to fit the policies and regulations you already operate under.
AI Workflow Assurance for regulated operations · inside your Trust Boundary · US & Europe
09 · Blog
Gartner expects most agentic AI projects to be scrapped by 2027. The failures are operational, not model quality, and four controls keep a project alive.
A dated, sourced map of AI agent rules in mid-2026: which EU AI Act and US state obligations apply today, which are delayed, and what to build for now.
Gartner expects 40% of agentic AI projects cancelled by 2027. The failures are operational, not model quality. Six documented patterns, the control for each, and a pre-mortem.
When a vendor retires or quietly changes the AI model your work depends on, the instructions you tuned, the tests you trust, and your costs can break overnight, with no change on your side. Here is how to stay ahead of it.