Adalma AI (@AdalmaAI) header
Adalma AI (@AdalmaAI) profile picture

Fraud and financial crime software for credit unions and community banks. Stop scams and elder exploitation before money leaves. Cernio · Cura · Tessra

Joined July 2026
Photo by Adalma AI (@AdalmaAI)
Adalma builds AI for financial crime compliance. The software runs in your environment. The agent assembles the case. A named officer signs it. Every filing lands on a tamper evident audit trail. Cernio covers AML, BSA and FinCEN e-filing. adalma.ai
1
2
93
Also applied to @speedrun SR008 with Adalma. Different format from YC: one team video, up to 3 minutes, at least half on the founder, the rest on product, market and progress.
1
1
253
Applied to @ycombinator W27 with Adalma this week. Solo founder. We build fraud and BSA software for credit unions, focused on the hardest case: older members exploited by someone they trust. A few things I learned writing it:
3
1
3
227
At the ABA Elder Fraud Prevention Summit today, Duke Han of USC walked through a decade of research on aging and money. Three things I'm taking home.
1
1
19
Vulnerable is not the same as exploited. A member who is struggling needs a call from someone who knows them. A member being drained needs a case.
1
3
The account often sees all of this before anyone else does. That is where we work. Thanks to @ABABankers for putting researchers, prosecutors and bankers in one room.
1
6
Most fraud tools are built to spot a stranger: a new device, an odd login, a payee nobody has seen. For most elder fraud, that's right. FinCEN's 2024 analysis found about four in five elder exploitation cases involved someone the victim didn't know.
1
8
A system tuned to flag unfamiliar people will pass a familiar one every time. The better question: who was just given access to this account, and what changed after? When was the POA added? Did contact details change? Did debits climb?
1
That data already sits in the core banking system. Most monitoring isn't set up to ask. Fraud and BSA teams: how do you catch exploitation by someone the member trusts?
Minnesota lets banks pause suspicious payments from older adults. In 2024: 638 investigations, holds placed in 19%, about $8M kept in accounts (DeLiema et al.). The law says "may." The hard part is spotting the case while the money is still there.
3
Regulators cleared banks to warn customers about suspected fraud on 2 Sept. FinCEN asked for scam-center SARs on 3 Sept. Detection is fine. Filing is fine. Nothing reaches the member before they hit send. Everyone is racing to write the SAR faster. The SAR is the receipt.
8
Pointed Jev at our own pipeline. Four typed questions in one request: does this page name a person, does that person sign off the ad budget, what is the direct route, is more research worth paying for. 205 to 484 ms a row. An unanswered question researches. It never skips.
5
risk = high (96%) manual review = yes (91%) that's the product. it's an AI-native if statement. instead of: if transaction > $10,000: review() you get: if "does this look suspicious?" > 95%: review()
1
10
"can it hallucinate?" it won't invent an option outside your list. but HIGH at 92% can still be wrong. constrained isn't the same as correct.
1
6
TypeSafe says it themselves: Jev is weak at reasoning, not trained on any domain. that's the design. it routes, it doesn't think. cheap fast decision layer in front of the expensive models. that's the interesting part.
5
LLMs generate answers. Jev makes decisions. sounds like a small difference. it changes the entire use case. instead of asking a model to describe risk, you define the allowed answers upfront and it returns one of them, with a probability.
1
8