I did a paralegal's job by hand for 3 years. Then I taught a machine to do it.
There is a moment every self-represented parent knows. You are standing at your kitchen table at two in the morning, surrounded by printouts, bank statements, message screenshots, and court forms you do not fully understand, and you realize that nobody is coming to save you. Not the lawyer you can no longer afford. Not the system, which was built by and for the people who work inside it. Nobody.
I lived inside that moment for years.
I spent 12 years dealing with the family court system. Ten of those years in litigation. Five of them with a lawyer, watching the invoices climb. Three of them completely on my own, because the money ran out long before the conflict did. And in those three years I learned something that changed the direction of my life: the work that costs families tens of thousands of dollars is not magic. It is preparation. Brutal, methodical, unglamorous preparation. And almost all of it can be systematized.
I know because I did it by hand first. Months of it. Organizing years of records into timelines a judge could actually follow. Cross-referencing every claim against a document that proved it. Learning the rules of court the hard way, one procedural mistake at a time. It nearly broke me, and I am a software developer with decades of experience building complex systems. I kept thinking about the parents doing this with no technical background, no time, and a child asleep in the next room.
So when the AI wave arrived, I did not see a toy. I saw the missing half of the machine I had been building in my head for a decade. Litigent AI is that machine. This is the story of what is inside it, why it is built the way it is, and what living on the wrong side of the system let me see that the industry would prefer stayed invisible.
The quiet truth about judges
Here is the insight that reorganized everything for me, and it is not cynical. It is the opposite.
A judge's job is to apply the law to the record in front of them. Not to your intentions, not to what really happened, not to the truth you carry around in your chest. The record. What was filed, what was served, what was entered, what complies with the rules of court, and what is supported by evidence.
Most self-represented parents lose ground not because the law is against them but because their record is a mess. They show up with feelings where the court needs facts, with piles where the court needs exhibits, with memory where the court needs documentation. And when the record is thin, the space fills with something else: reputation, familiarity, the comfortable shorthand of professionals who have worked the same courtroom together for twenty years.
But flip that around and something powerful happens. When a self-represented parent walks in with a record that is complete, organized, compliant with the rules, and supported line by line by real documents, the room for all of that shorthand shrinks. The judge is looking at a clean record and the law that governs it. That is exactly what the court is designed to do, and it is exactly what a prepared parent should want. You are not asking anyone for a favor. You are narrowing the case down to the thing the court is actually obligated to honor: the rule of law, applied to a record you built properly.
Lawyers understand this deeply. It is the core of what you are paying for. The uncomfortable part is what it costs to get it, and how the billable hour makes thoroughness and expense the same thing. Every document reviewed, every timeline assembled, every draft revised is metered time. The industry has no structural incentive to make preparation cheap, because preparation is the product. I do not say that with hatred. I say it as someone who paid for it for five years and then learned to do it myself for three.
That preparation layer, the expensive, mechanical, meterable layer, is what we built Litigent AI to collapse.
What we actually built
I will not hand over the blueprints. But I can tell you what the system does and the standards it is held to, because those standards are the entire point.
It starts with capture. A family court case is a swamp of raw material: years of messages, financial records, photos, incident notes, school emails, medical letters, prior orders. The first thing our tooling does is turn that swamp into structure. Documents are ingested and read page by page, communications are categorized, events are placed onto a timeline, and every extracted fact stays pinned to the source it came from. Nothing floats free. If a figure appears in a draft, it is because a named document put it there, not because a model guessed.
Then comes the vetting. This is where I put the scar tissue. Every single AI call in our platform runs through one instrumented pipeline that records what was asked, what was produced, how long it took, and what it cost, so that quality is measurable instead of anecdotal. We maintain a growing library of reference cases with known correct answers and replay our systems against them, scoring the results across dimensions like groundedness and faithfulness: is the answer supported by the sources, and does it stay inside them? Releases are gated on those scores. If a change makes the drafting or analysis worse, it does not ship. We even run shadow comparisons on document transcription, measuring agreement between engines so accuracy is a tracked number, not a hope.
Then the jurisdiction layer. Family law is not one thing. It changes province by province, country by country, form by form, deadline by deadline. The system carries dedicated jurisdiction packs so that what it produces is shaped for the rules of court where you actually are, in the language those courts expect. A brilliant document in the wrong format is a rejected document. I learned that one personally, and it cost me months.
And underneath all of it, the rails. This is the part I am proudest of, and the part most people never see. Litigent AI is an information and preparation tool. It is not a lawyer, and it must never quietly pretend to be one. So we built the boundary in triplicate. The scope-of-service line is engineered into every agent at the system level. A second, independent layer reviews candidate outputs and rejects anything that drifts from information into advice. A third runtime gate hard-blocks outcome promises, because nobody can guarantee a result in litigation, and any tool that says "you will win" is lying to a desperate person. And before any of that, a safety gate watches for crisis. If someone is in danger, the system stops being a legal tool entirely and routes them to real human help. That is not a feature we advertise. It is a feature I insisted on because I know what the worst nights of this process feel like.
The privacy architecture, because this data is sacred
The material in a family court case is the most sensitive data a person will ever hold. Allegations. Finances. Children. So privacy is not a settings page for us. It is architecture.
Sensitive fields are encrypted at rest with the same class of encryption used in serious financial systems. Our AI telemetry is built on cryptographic fingerprints, so we can measure quality and reproduce results without warehousing the plain text of your case in an analytics system. Identifying details are stripped from context before it ever leaves for processing, with tiered privacy modes that let a user dial that all the way up to keeping their material out of cloud processing entirely. Your content is never recycled into our evaluation library without your explicit opt-in, a line we drew to match Canadian privacy law principles, and the default answer is no. Audit records expire on a published schedule instead of living forever. When you share something with a mediator or another party, the link is expiring, revocable, watermarked, and access-logged, because handing someone a copy of your case should never mean losing control of it. And when a document needs redaction before filing, the redaction is permanent replacement of the text, not a black box drawn over words that are still underneath.
None of this is glamorous. All of it exists because I kept asking one question while we built: what would I have needed to trust this with my own case? Then I made that the standard, because it was my own case, for ten years.
What this means for a family
Let me make it concrete.
The work I did by hand across months, the timeline construction, the document organization, the disclosure preparation, the drafting and revising against the rules of court, is work our tooling now does in a fraction of the time, for a fraction of what I spent. The parent using it walks into a lawyer's office, if they use one at all, with their material already structured, already sourced, already in order. Every billable hour they do buy is spent on judgment and strategy, the things genuinely worth paying a professional for, instead of on sorting paper at several hundred dollars an hour.
And the parent who cannot afford any of it, the parent I was in those last three years, gets something that did not exist when I needed it: a way to show up prepared. Not a lawyer. Not a promise. A fighting chance to put a clean, compliant, evidence-backed record in front of a judge and let the law do what it is built to do.
That is what the industry's structure never had a reason to give you. I did not build this to be liked by the profession. I built it because families are draining their savings, their equity, and their sanity into a conflict machine during the exact years they should be pouring everything into their kids. Both homes end up poorer. The children inherit the stress. Multiply that across every courthouse in every province and state, every year, and you are looking at damage to the foundation of a society, done quietly, one retainer at a time.
Why I will not stop
People ask me why I would spend my software career on family law, of all things. The honest answer is that I did not choose it. It chose me, one court date at a time, for twelve years.
I got through. Not unscarred, but through, and with a rare combination on the other side: deep lived knowledge of what this process demands, and the engineering ability to build the machine that meets those demands. Very few people have both. So the way I see it, I do not really have the option of doing something else with them.
Litigent AI exists so that the next parent at that kitchen table at two in the morning is not alone with a pile of paper and a deadline. It exists so that preparation, the great equalizer in a courtroom, is no longer a luxury good. It exists because mothers, fathers, and children everywhere deserve a process that costs them less than their future.
We are starting in New Brunswick, where I lived every page of this story. We are building for every parent, everywhere, who has to walk in alone.
You are not alone. Not anymore.
Litigent AI is an AI legal information and preparation toolkit. It is not a law firm or a lawyer, and nothing in this article or in the product is legal advice. For advice on your specific situation, speak with a lawyer, legal aid, or the court's self-help services in your jurisdiction. What we will do is help you walk in ready.