BRCI — Banca Română de Credite și Investiții × Carol Calin
Confidential proposal · prepared for BRCI

Your clients' money sits in a vault.
Where does their data sit?

You manage accounts, credit and investments with rigour and under banking secrecy. I propose to bring that same discipline to how BRCI adopts artificial intelligence — keeping every piece of client data inside the bank.

For  The executive management, BRCIBy  Carol Calin · Data Scientist & AI BuilderDate  June 2026 · Bucharest ⟷ Amsterdam
The situation

The AI shift is here. For a bank, the real question isn't whether — it's how, without breaking banking secrecy.

As a credit & investment institution supervised by ASF and the National Bank of Romania, your edge is trust: banking secrecy, compliance, and a relationship built over time with each client. That is precisely why the obvious AI tools are dangerous to you — and why the right approach becomes a genuine advantage.

I.

The wave is real

Reporting, KYC, onboarding, compliance memos, client correspondence — the back-office work that fills your week is now largely automatable. Banks that adopt well will quietly run leaner and respond faster.

II.

The confidentiality trap

Pasting a client's data or a credit file into a public chatbot ships privileged data to a third-party cloud. For a bank bound by banking secrecy and GDPR, that is the fastest way to turn a productivity tool into a compliance incident.

III.

The asymmetric opportunity

You don't need a 30-person IT department. With AI agents, one focused builder can ship the bespoke internal tools an agile bank needs — on infrastructure you control. Small and private becomes faster, not slower.

What I would build with you

Five workstreams, mapped to the back-office you already know: KYC · reporting · onboarding · compliance · risk.

Each is something I have already built in a harder setting — on sensitive biometric data, forensic evidence, or under cryptographic constraints. Here is how that transfers to BRCI.

01 — Confidentiality first

Private & local AI — your data never leaves the bank

Deploy open-weight models (Llama, Mistral, Qwen) on hardware you own, plus privacy-preserving techniques so analysis happens without raw client data ever being exposed. The default, not an afterthought.

Why it fits BRCIBanking secrecy & GDPR stay intact; nothing is sent to a US cloud. Confidentiality becomes a feature you can show clients and regulators.
I've done the hard versionTrained ML on private biometric data using homomorphic encryption (PySyft, TenSEAL) and zero-knowledge ML (EZKL) — proving results without revealing the underlying data.
02 — KYC & anti-fraud

Fraud, forgery & deepfake detection for onboarding

Automate document checks and flag manipulated IDs, statements, and the new wave of deepfake video/voice used to defeat remote KYC and impersonate executives.

Why it fits BRCIOnboarding and KYC are among your friction points — and deepfake-enabled fraud is now hitting banking and investment firms directly.
I've done the hard versionDeepfake-detection research at the Netherlands Forensics Institute; thesis defended at the Dutch Ministry of Justice & Security, informing national forensic strategy.
03 — Back-office & reporting

Agentic automation for the work that drains hours

Consolidated reporting, drafting of compliance memos, position summaries and correspondence — generated, then reviewed by your team. Humans approve; agents do the typing.

Why it fits BRCIReporting and back-office are on your list of hard spots. Less manual stitching, faster turnaround, fewer copy-paste errors.
I've done the hard versionI ship end-to-end systems solo using AI coding agents (Claude Code, Cursor, Devin) — out-building larger teams across 10+ international hackathons.
04 — Knowledge indexing

A private knowledge copilot — the bank's "conductor"

Index your BNR/ASF regulations, MiFID rules, internal procedures and (permissioned) files into a private assistant that answers with citations. Your advisors ask in Romanian; it replies from your documents only.

Why it fits BRCIAn agile bank's expertise lives in a few key heads. Indexing it makes every advisor faster and protects you when someone is away.
I've done the hard versionBuilt a multi-agent shared-knowledge base with concurrency control and semantic indexing for coordinating AI agents over a private document store.
05 — Enablement · "token capital"

Train the team — and make your AI spend efficient

The best tool is useless if the people in the branches and at head office don't trust it. I run hands-on sessions (in Romanian or English) that turn advisors, back-office and support into confident users — building reusable "skills" and prompt playbooks for your real tasks. And I treat your AI budget like what it truly is: capital to allocate, not a flat subscription. (More on that below.)

Why it fits BRCIYou're a relationship business. A psychology + community background means I teach to people, not to engineers — adoption that actually sticks.
"token capital" efficiency

You allocate capital for a living. AI tokens are just another asset class to allocate well.

Most firms either avoid AI or burn money sending every task to the most expensive frontier model in the cloud. Neither is prudent management. I treat tokens and compute the way you treat a mandate:

Route by value. Routine work runs on cheap local models (zero marginal cost, full confidentiality). High-stakes judgment is reserved for frontier models — used sparingly, with no client data attached. Cache and reuse. Knowledge indexing means you don't re-pay to "re-read" the same documents. Measure ROI. Every workflow gets a cost-per-task and a time-saved figure, reported to you like a position.

Illustrative allocation
Routine drafting, summaries, KYC pre-checksLocal · €0/req
Sensitive client data — anythingOn-prem only
Complex reasoning, rare & de-identifiedFrontier · metered
Repeated document lookupsCached
Reporting line to managementCost & time saved
Why me

I'm not a generalist who'll plug you into someone else's cloud. I build verifiable, private ML — and I ship.

I

Forensic ML at a Ministry of Justice

Fine-tuned vision transformers to detect manipulated video at the Netherlands Forensics Institute; thesis defended at the Dutch Ministry of Justice & Security. Fraud & forgery is my actual research domain.

II

Privacy-preserving by training

Homomorphic encryption (PySyft, TenSEAL), federated learning, and zero-knowledge ML (EZKL). I make models work without exposing the sensitive data underneath — exactly your constraint.

III

One builder, team-scale output

Solo founder shipping complex systems with AI coding agents. Finalist / winner across ETHGlobal, EigenLayer, NEAR and more — speed is the whole point.

IV

~$30k in non-dilutive grants & prizes, solo

Including a 3rd-place finish (12,125 participants) in a challenge sponsored by Vitalik Buterin, and a research collaboration with Vanderbilt University. I deliver outcomes, not slide decks.

Plain and honest: I have not worked inside a bank, and I won't pretend otherwise. What I bring is the part that's genuinely hard to hire — private, verifiable AI on sensitive data, forensic-grade fraud detection, and the speed to ship — paired with a willingness to learn your business from your team. The domain I'll learn from you; the technology I already know cold.
A low-risk way in

No big-bang transformation. We prove value on one workflow, then scale only what works.

01
Step one · ½ day · complimentary

Confidential AI & data audit

A short on-site or remote session with your team. We map where AI helps, where confidentiality forbids the cloud, and pick the single highest-value workflow to pilot. You leave with a written plan — whether or not we continue.

02
Step two · ~3 weeks · fixed-fee pilot

One workflow, fully working, fully private

I build the chosen pilot — most likely the private compliance/reporting copilot — running on infrastructure you control. Clear success criteria agreed up front. Small, bounded, and yours to keep.

03
Step three · ongoing · only if it earns it

Scale & enablement

Roll the proven pattern across KYC, onboarding and back-office; train the team; set up your "token capital" routing and reporting. Retainer or per-project — your call, reviewed quarterly like any mandate.

Letter of intent
BRCI
Bucharest · Str. Ștefan cel Mare nr. 3
June 2026
Confidential

For the attention of the executive management of Banca Română de Credite și Investiții (BRCI).

Dear Sir or Madam,

I came across BRCI as a credit & investment institution on the ASF register — a Romanian bank combining lending, savings and investment services for retail and corporate clients. That is exactly the environment I work best in: precision, discretion, and results you can verify.

I propose a clear engagement: to help BRCI adopt artificial intelligence without ever compromising your clients' confidentiality — local models, fraud and forged-document detection, back-office and reporting automation, and a team trained to use it with confidence. No opaque cloud, no data leaving the bank.

This page is not a generic pitch: I designed it specifically for you, in your colours, from your own priorities. If the approach resonates, I suggest we start small — a half-day audit, with no commitment — to identify together the first, highest-value workstream.

The engagement can be delivered in Romanian or English. I am at your disposal for a 30-minute conversation.

Carol Calin
Data Scientist · AI & Privacy-Preserving ML · Amsterdam
Get in touch

Let's talk. 30 minutes, no obligation.

A 30-minute, no-obligation conversation. I'll come prepared with two or three concrete ideas specific to BRCI.