• Software Engineer, New Delhi. 3+ years building the production systems behind a trading desk: a distributed transaction platform, a ZeroMQ microservice engine, high-volume data pipelines, dashboards and internal tools.
• Platform SDK serving 150+ connected clients with ₹400+ Cr ($48M) in daily volume across three broker integrations (Zerodha, Interactive Brokers, XTS): real-time validation, reconciliation, P&L and audit trails. Eliminated 100% of manual workflows.
• Performance engineering: rebuilt a data-intensive batch engine from about 150 hours to under 1 hour with NumPy vectorization, O(log n) lookups and 16-way parallel orchestration.
• Public projects: author of pyzdata (open-source library, CLI and Streamlit app on PyPI) and KiteWeb (source-available), 60+ stars across projects.

pip install pyzdata

Download OHLCV and open-interest data for any NSE, BSE, NFO or MCX instrument, from 1-minute to daily. Ships as a Python library, a CLI and a Streamlit web app. Parallel downloads, disk caching, rate limiting, retries and a typed exception hierarchy. Package engineering: 90 unit tests, PEP 561 typing, Python 3.10 to 3.13, and trusted publishing through GitHub Actions with Sigstore-signed releases.

Python PyPI CLI Streamlit Threading Pandas pytest GitHub Actions

Unofficial Python client for Zerodha Kite's web endpoints: authentication, session management, and order and market-data calls. Built for learning and personal automation, and adopted across the Indian retail algorithmic-trading community. Source-available for personal use, not affiliated with Zerodha.

Python REST Requests Zerodha

Private code, described in the resume site with diagrams.

• Multi-broker transaction platform. Python SDK automating the full transaction lifecycle across Zerodha, Interactive Brokers and XTS: signal intake, order placement, real-time risk checks, position reconciliation, P&L and end-to-end audit trails. Serves 150+ client accounts with ₹400+ Cr in daily volume.
• ZeroMQ microservice engine. A central broker holds the single upstream session while five independent worker processes communicate by asynchronous message passing. CSV-backed state machines give crash recovery and idempotent restarts. Runs autonomously on AWS EC2 with IBC-managed restarts, driving US options strategies end to end.
• pgcbacktest, a 150× faster batch engine. NumPy-vectorized engine with O(log n) lookups and a FileLock orchestrator distributing jobs across 16 parallel workers; a Polars and DuckDB aggregation layer feeds Streamlit dashboards and Excel exports. 50,000+ job combinations per cycle.