TechDesk

Algo Trading Platform

Automated strategies, your broker, your control

Industry
Fintech · Trading automation
Status
Live
Our role
Design, engineering and operations
Services
AI & Intelligent Automation, Web & Product Engineering, Software Engineering

Overview

A cloud algorithmic-trading platform: build strategies without code, backtest them, paper trade by default and go live through the trader's own broker, with AI research running alongside.

The problem

  • Manual execution is slow and emotional; active traders cannot watch charts all day.
  • Testing an idea usually means writing code, and going live too early is costly.
  • Equity and crypto usually live in two separate worlds.

The approach

One Strategy Builder where traders compose rules from forms and settings, then backtest, paper trade and only then go live, with explicit consent. The engine runs on the server during market hours, so the trader's laptop does not have to.

System architecture

  1. 01

    Strategy Builder

    Point-and-click rules: indicators, opening range, risk settings

  2. 02

    Backtester

    Historical simulation on any stock, index or index basket

  3. 03

    Execution engine

    Fetches candles, evaluates signals, monitors stop-loss and targets every few seconds

  4. 04

    Broker adapters

    Angel One (Indian derivatives) and Delta Exchange (crypto, 24/7)

  5. 05

    AI layer

    Live Research on a 5-minute cycle and Swing Trade AI, with multi-provider LLM failover

  6. 06

    Trust layer

    Paper by default, recorded live consent, audit log, encrypted broker tokens

Technology

  • React
  • TypeScript
  • Vite
  • TanStack Query
  • Node.js
  • MongoDB
  • AWS Amplify
  • Groq · NVIDIA · OpenRouter LLMs

Key features

  • No-code Strategy Builder with backtesting before any live order
  • Paper trading by default; live trading only after recorded consent
  • Trade guards, cooldowns and database-level duplicate-trade prevention
  • AI Live Research cards with entry, stop-loss and targets, placeable in one click
  • Swing Trade AI with a dedicated position monitor
  • Telegram and WhatsApp alerts; audit log for logins, consent and mode changes
  • Starter, Pro and Ultra Pro plans billed monthly, half-yearly or yearly

Challenges solved

Model layers that fail silently

A single LLM provider can go down or rate-limit without warning. The AI layer runs an environment-driven provider chain with per-attempt timeouts, an overall deadline, failure classification and per-provider cooldowns, so one outage degrades a feature instead of killing it.

Duplicate or runaway trades

Guards live in the database, not just in code: partial unique indexes make a second identical order impossible even under concurrent signals.