our work
Case studies
These are real projects TRAGenX has built — financial intelligence, compliance workflows, security platforms, and trading systems. Each one is framed honestly: the problem it set out to solve, how we approached the build, and what it actually delivers. No inflated metrics, no invented logos. Read a case, then open the full product detail.
FinSage
AI-Powered Financial Intelligence
- Problem
- Compliance teams drown in manual AML review of financial documents.
- Approach
- An AI-powered AML platform that extracts transactions and scores risk across pattern + behavioral categories, with KYB/UBO entity intelligence and a Wolfsberg CBDDQ form engine.
- Outcome
- Structured, regulation-cited AML findings from uploaded documents — traceable and audit-ready.
- Python 3.13 + FastAPI (hexagonal)
- React 19 + Vite
- Tailwind + shadcn/ui
- Zustand
- PostgreSQL 15 (pgvector, pg_audit, pg_partman)
- Google Gemini 2.5 Flash
- MinIO / S3
- Celery + Redis
- Business-registry integrations
- SMTP
TradegenX
Claude-Driven Trading Orchestration
- Problem
- Running automated trading with an LLM safely — without hallucinations moving real money.
- Approach
- A Claude-orchestrated layer where every action passes deterministic safety gates (leverage/position/duration caps, backtest + walk-forward gates), driven from a CLI/daemon + Telegram.
- Outcome
- A positive-expected-value, drawdown-controlled, auditable system where the LLM is advisory only.
- Python 3.11+
- typer + rich CLI
- httpx
- pydantic v2
- structlog
- APScheduler + aiosqlite daemon
- python-telegram-bot
- systemd (local, no Docker/cloud)
- Claude (Sonnet ticks / Opus propose + review) via headless claude -p
- MCP market-intel tools
- Sitragenx Portal (VCPy) REST API
Sitragenx
Automated Futures Trading Engine
- Problem
- Designing and validating trading strategies safely needs design, backtest, and execution in one place.
- Approach
- A platform unifying a visual strategy builder, a candle-by-candle backtester, hyperparameter optimization, and live bot execution.
- Outcome
- A production system with per-candle decision auditability; live trading is explicit opt-in.
- Python 3.12 + FastAPI
- async SQLAlchemy 2.0
- Celery + Redis (RedBeat)
- Next.js + React
- shadcn/ui + React Flow
- PostgreSQL (+ TimescaleDB)
- Optuna
- NumPy + TA-Lib
- structlog + Prometheus + Sentry
- Binance Futures
PyramidOS
The Operating System for Pentest Firms
- Problem
- Pentest firms juggle fragmented tools across the whole engagement lifecycle.
- Approach
- A multi-tenant SaaS spanning lead capture, project management, findings library, QA, branded reports, and a secure client portal.
- Outcome
- One platform from lead to delivered report, with quality gates and tenant isolation.
- Django 6 + DRF
- Next.js 16
- PostgreSQL (row-level security)
- Redis + Celery
- MinIO / S3
- WeasyPrint + pikepdf (AES-256)
- Google / Microsoft OAuth2 + JWT + 2FA
- Stripe
- django-fsm-2
Fin-On
Compliant B2B Onboarding Workflow
- Problem
- EU/Lithuania-regulated customer onboarding is complex and easy to get wrong — steps get skipped.
- Approach
- A digitized Wolfsberg CBDDQ intake with parallel registry enrichment, KYC for every UBO, rule-based risk scoring, and role-based approval.
- Outcome
- An onboarding workflow where no compliance step can be skipped — bilingual and audit-ready with a 7-year trail.
- Python 3.12 + FastAPI
- async SQLAlchemy 2.0 + asyncpg + Alembic
- Next.js 16 + React 19
- PostgreSQL 17
- Redis + Celery
- Microsoft Entra ID SSO
- WeasyPrint
- Google Cloud Storage (AES-256)
- SSE + Redis pub/sub
- Cloudflare WAF
Fin-Score
Automated SME Loan Decisioning
- Problem
- SME credit decisions take days of manual underwriting.
- Approach
- Parallel data enrichment, configurable knockout rules, and a 3-category weighted score driving a fully automatic decision.
- Outcome
- Approve/reject decisions in minutes within a defined envelope — every decision fully auditable.
- Python FastAPI
- async SQLAlchemy 2.0 + Pydantic v2
- Next.js 16 + React 19 + Tailwind 4
- Celery + Redis
- PostgreSQL
- WeasyPrint + Jinja2
- Microsoft Entra ID SSO
- Smart-ID / Mobile-ID signature
- Google Cloud Storage (AES-256)
- Cloudflare
- hexagonal ports-and-adapters
SalesGenX
Your Outbound Sales Team, Run by AI Agents
- Problem
- Outbound sales doesn't scale without a floor of SDRs, and generic AI cold email gets ignored or flagged as spam.
- Approach
- We turned our own group's go-to-market into an autonomous outreach engine: agents source and research prospects, write human-reading, on-brand cold emails, and pass each one through deterministic anti-AI-tell guards, a blind adversarial reviewer, and per-region compliance before a single human persona sends — then agents triage replies and book meetings. A deterministic brain plans each day within hard, non-bypassable caps, with an always-on kill-switch.
- Outcome
- A complete outbound department that runs itself end to end — sourcing to booked meeting — with humans setting the guardrails. Live proof that we can turn a real operation into safe, supervised AI agents.
- Claude (Sonnet) agents
- Python
- PostgreSQL
- Deterministic agent orchestration
- SMTP / IMAP email
- Cal.com booking
- Per-region compliance engine
get in touch
Have a project like these?
If one of these resembles what you need built — or you want to talk through something new — email is the best way to reach TRAGenX. Send us a note and we'll get back to you.