Selected work

Multi-agent orchestration system using Google Vertex AI (Gemini) and Claude as intelligent orchestrators — with DAG-based task routing and fallback model chains applied to trading strategy research and signal analysis.

• –DAG-based task routing with fallback model chains across Claude, Gemini, OpenAI, and local models via Ollama.
• –Cost-aware model routing with subagent pipelines for parallel workloads.
• –CLI harnesses for multi-provider LLM access with structured output parsing.

Fully automated trading engine with three decoupled modules — Data Adapter, Position Manager, and Executioner — using Kotak Neo APIs and AMQP-based messaging for distributed real-time execution.

• –Three-module architecture with async AMQP messaging for decoupled execution.
• –Logic-based portfolio manager handling individual and paired trade management.
• –Real-time data adapter feeding normalised tick data to downstream modules.

Multi-broker unified execution backend supporting NSE and NYSE markets across IIFL, SMC, Zerodha, and 5Paisa — with asynchronous APIs, high-speed data routing, and webhook-triggered order execution.

• –Unified broker abstraction layer across four brokers (NSE + NYSE).
• –Webhook-triggered order execution backed by Redis state management.
• –WebSocket / Socket.IO publish-subscribe system for live trade distribution.

Trading system driven by TradingView webhook events — receives signal payloads, executes buy/sell orders, and uses Redis for distributed state management and fault tolerance across concurrent sessions.

• –Webhook listener processing TradingView signal events into live order execution.
• –Redis-backed state machine for fault-tolerant position and session tracking.
• –Designed for reliability under concurrent multi-account order flows.

End-to-end ETL pipeline for ingesting and processing client trading contract notes — parsing raw documents, transforming into structured records, and exposing clean data via RESTful APIs.

• –Automated parsing and transformation of trading contract note documents.
• –RESTful API layer for downstream data access and integration.
• –Long-running pipeline maintained across 2+ years of production operation.

Backtesting and optimization framework for evaluating algorithmic trading strategies — running simulations against historical data and surfacing performance insights through data visualization.

• –Strategy simulation engine running against historical OHLC and tick data.
• –Optimization loops for parameter tuning across strategy variants.
• –Data visualization layer for backtest result analysis and reporting.

Production trading system for executing live orders on the NSE — built with a concurrency-first design using multithreading and asynchronous I/O to handle real-time market data and order flow without latency bottlenecks.

• –Live order execution on NSE with multithreaded request handling for throughput.
• –Asynchronous architecture for real-time market data consumption and order dispatch.
• –Designed for reliability under concurrent multi-instrument trading scenarios.

Automated execution engine that trades on pre-defined market signals — combining a Flask-based API layer, multithreading for concurrent order management, and WebSocket feeds for low-latency market interaction.

• –Signal ingestion and order execution pipeline with configurable strategy parameters.
• –Flask API for runtime control and monitoring of active bot sessions.
• –WebSocket integration for real-time market data and trade confirmation feeds.

Unified data platform consolidating structured and unstructured data from multiple disparate sources into a single, query-optimised database — with a schema designed for analytical performance and reporting.

• –Designed and implemented a unified schema to normalise data across heterogeneous sources.
• –Built ingestion pipelines to pull, transform, and load data from multiple upstream systems.
• –Optimised database layout for query performance and downstream reporting workloads.