Our CTO Oleksii Koshlatyi on how the technology works, why every case makes it sharper, and where it’s heading.
Most companies building AI for blockchain investigation start with a model and look for data to feed it. Recoveris started the other way around. The AI behind Recoveris was built by forensic investigators, for forensic investigators, so the people who run real cases are the same people shaping the technology. That inversion changes what the AI can do, how it’s trained, and how much a client can trust the output in front of a regulator or a court.
Why is the AI built by investigators instead of engineers alone?
Recoveris was built by forensic investigators, for forensic investigators. It’s an easy line to write and a hard thing to do, because it means the corpus that teaches the model is real casework rather than scraped public data. As Oleksii puts it, the investigators are the engine behind the AI, and the people who run real cases are the same people shaping the model. That isn’t a slogan. It’s how the models are trained.
Right now the company is in what Oleksii calls the Foundation stage, where the cases themselves are the corpus. The AI has already been trained on hundreds of real cases and thousands of transactions, and it keeps learning from every new investigation the team runs.
How does every case make the AI sharper?
Every case Recoveris handles becomes training data. When an investigator approves or rejects a suggestion from the AI, the models are retrained on that decision, so every day of casework makes the system a little sharper. Oleksii calls that feedback loop the company’s moat, and the reasoning is simple. The cost of building software keeps falling every year. Proprietary data from real investigations does not. Every case Recoveris runs feeds a flywheel that makes the next case faster, and that compounding advantage is very hard for a competitor to copy, because they’d have to run the same volume of real cases to build the same corpus.
This is the core information-gain point of the whole approach: the value doesn’t sit in the code, it sits in the labelled decisions of working investigators, accumulated case by case.
What does the AI actually do today?
Today the AI works as a co-pilot. It triages the data, helps build the transaction graph, and explains in plain language why a transaction looks illicit. It does this on two sides. In the front office, the Case Intake Platform, it helps victims submit a clean, structured case. In the back office, it helps investigators move faster through the trace. Throughout, the investigator always stays in control. The AI proposes, the human decides, and that decision feeds back into training.
Where is the roadmap heading?
The roadmap has three stages. Stage one, today, is human-led with the AI assisting. Stage two, this year, is the co-pilot working alongside the investigator on live cases. Stage three is the Digital Investigator, where the AI leads the investigation and a human oversees it. The North Star, in Oleksii’s words, is a fully functioning AI investigator.
Why is investigation volume creating a whole new category?
The volume problem is what makes autonomous investigation a real category rather than a feature bolted onto existing tools. Oleksii cites a single number that captures it: Binance alone received 71,000 law enforcement requests in a single year. No human team scales to that. The requests keep coming, the caseloads keep growing, and manual tracing cannot keep pace. Recoveris is building toward that scale one case at a time, which is the only honest way to build an AI investigator you can actually trust.
What sets the Recoveris approach apart?
Three things set it apart, and Oleksii is direct about each one.
First, vendor neutrality. Chainalysis, TRM Labs, and Merkle Science are building AI investigators too, but each one is locked to its own data. Recoveris connects the tools investigators already use and cross-checks between them, which gives a more complete picture than any single dataset can produce on its own.
Second, the model is a white box. It’s fully deterministic, so Recoveris can show exactly why a transaction was flagged. That matters in court, and it matters for compliance teams that have to justify every decision to a regulator. A black box that produces a verdict without a traceable reason is a liability in both settings.
Third, ownership of the full loop. Nobody else in blockchain investigation owns the full path from investigation to recovery, and Recoveris does. That end-to-end control is what turns a flagged transaction into a frozen and recovered asset.
How does Recoveris keep client data safe?
Recoveris treats client data as the most sensitive thing it holds, and the safeguards sit in the architecture from day one. Start with the AI: the live system never touches a public language model. Everything runs in a closed environment on Recoveris’s own servers, and client data never leaves to train someone else’s system. A public model was used once, only as a teacher, and the live system is fully self-contained.
Then there’s the geopolitical layer, which Oleksii expects to matter more every month. European clients don’t want a single US analytics company holding all their data, and that concern is only growing. At that point, vendor neutrality and full data control stop being nice-to-have features and become mandatory. Because the model is a white box, every decision is traceable and auditable, so data safety is built into how Recoveris stores, isolates, processes, and accounts for every case.
The takeaway
The bet Recoveris is making is that in AI-driven blockchain investigation, proprietary case data beats raw model size, and a white-box that a court can audit beats a black box that a court cannot.
If your organisation or users have been affected, or you want to test the tools Oleksii describes, schedule a confidential consultation with Recoveris.
