AI agents. Rust backends. Fintech infrastructure. KYC/KYB/AML operations.
No demo theatre. No black-box compliance fantasies. Just systems that ship, scale, and survive reality.
I design AI-native systems where models prepare, humans validate, and audit trails prove.
My work sits at the intersection of:
- Production GenAI and agentic workflows
- Rust backend systems with strong domain models
- Fintech, KYC, KYB, AML, and LCB-FT operations
- Human-in-the-loop automation for regulated environments
- Venture studio execution from idea to production
The thesis is simple:
AI will not replace serious workflows until it becomes typed, observable, permissioned, and auditable.
That is what I build.
The first wave of AI agents was fragile Python scripts wrapped in hope.
The next wave will be:
- typed
- async
- observable
- permissioned
- auditable
- connected to real business systems
- safe enough for fintech and compliance operations
My current stack leans heavily into:
Rust · Axum · Tokio · Postgres · OpenTelemetry · LLM orchestration · RAG · structured outputs · human validation
| Project | What it is |
|---|---|
| Remolab | Venture studio infrastructure for AI, fintech, and deep-tech products. I help teams turn vague ambition into shippable systems without drowning in architecture debt. |
| Welcome Place | Fintech and operational infrastructure for migrants and newcomers in Europe, including regulated onboarding, KYC/KYB workflows, and product architecture. |
| CaseReady / RavenKYC | Supervised AI assistant for blocked KYC, KYB, AML, and LCB-FT cases. AI prepares the work. Analysts validate. Audit trails prove what happened. |
| AI Reading Club | Public learning space for foundational LLM papers, from transformers to interpretability, with practical engineering discussion. |
| Rust AI Systems | From-scratch ML and transformer explorations in Rust, because apparently reading papers was not painful enough. |
- schema-first AI outputs
- deterministic workflows
- evaluations and prompt regression tests
- tracing, observability, and cost control
- analyst validation instead of blind automation
- boring reliability in systems that cannot afford drama
Because the model can be magical.
The production system still needs to behave like adults built it.
I run the AI Reading Club, a GitHub-based reading club for foundational AI and LLM papers.
Topics include transformers, attention, BERT, generation, interpretability, scaling, fine-tuning, and implementing ML ideas from first principles.
Repository:
github.com/hghalebi/ai-reading-club
I am interested in serious conversations around:
- production GenAI
- regulated AI workflows
- fintech infrastructure
- Rust backend architecture
- KYC / AML automation
- agentic operating systems
- venture studio execution