NinjaNext product
Cognix: a trading signal and research platform
Our in-house platform for generating, testing and delivering trade-signal setups across crypto, forex and binary markets, with a validation pipeline and an LLM idea agent.
Runs in production on our own server. It is an internal tool with no public sign-up.
- Owner
- NinjaNext product
- Industry
- Fintech & Trading Tools
- Period
- 2026
- Platforms
- Admin web console, Backend service, Telegram
- Key figures
- 3markets on one shared core
- 8validation stages before a strategy goes live
Cognix is shown as an engineering project. We make no claims about trading performance. Nothing here is financial advice, and trading carries risk.
The problem
Trading strategies are easy to over-fit and hard to judge honestly. We wanted a system where every strategy change is measured after fees and slippage before it is used, every setup and its result is recorded, and risk limits are enforced by the software rather than by memory.
What we built
Cognix is a Kotlin and Spring Boot backend with a React admin console. It stores one-minute market data, runs strategy modules for three markets on a shared core, backtests changes net of costs, and moves strategies through research, validation and paper trading before a person can promote them. Setups are delivered to Telegram in formats that common trade-copier tools read.
Key features
- Strategy modules for crypto futures, forex and binary markets on a shared core of indicators and a position tracker
- One-minute candle store with bulk imports, so live runs and backtests read the same data
- Backtests net of fees and slippage, each saved with a snapshot of its settings
- An eight-stage validation pipeline including walk-forward tests and a 14-day paper-trading period
- Strategy lifecycle from researched to validated, paper and active, with promotion by a person only
- Nightly automated research loop that screens candidate ideas in several stages
- Daily scheduled Claude agent that proposes one strategy hypothesis for the research loop
- Risk controls: caps on open setups, cooldowns, a daily loss guard and a log of every admission decision
- Telegram delivery with threaded follow-ups and daily reports, plus TradingView webhook input
- Admin console with dashboard, exposure, strategies, backtests, engines, audit log and backups
- Deploys through GitHub Actions with a health check and automatic rollback
Architecture and stack
Apps people use
- Admin web console
- Telegram
Backend and business rules
- Kotlin
- Spring Boot
- Spring Security + JWT
- PostgreSQL
- Flyway
- Bucket4j
- Python
Connected services
- Claude (Anthropic)
Full stack
- Kotlin
- Spring Boot
- Spring Security + JWT
- PostgreSQL
- Flyway
- Bucket4j
- Python
- React
- TypeScript
- Vite
- Ant Design
- TanStack Query
- Zustand
- Recharts
- Vitest
- Claude (Anthropic)
- GitHub Actions
Our role
Designed, built and operated in-house by NinjaNext.
Challenges and how we handled them
Measuring changes honestly
A strategy that looks good before costs often fails after them. Every backtest includes fees and slippage, uses separate training, validation and test periods, and is stored with the exact settings that produced it.
Automation with a person in charge
The research loop and the LLM agent suggest and test ideas on a daily schedule, but only a person can promote a strategy to live use, and only after it has passed validation and paper trading.
Safe releases for a system that never sleeps
Each deploy runs a health check against the new version and rolls back on its own if the check fails, and a notification reports what happened.
Logo artwork only. We do not publish screenshots of trading data.
Let's build it
Planning something similar?
Tell us about your project. If it is close to Cognix, we will explain what we would reuse, what we would do differently and what to decide first.