Shreejit Verma

• Built low-latency components of the automated market-making stack for the Prime Credit Market (average $500M of daily market-making volume), spanning real-time market-data ingestion, tick analytics, and pricing/execution paths.
• Profiled and optimized the software hot path feeding FPGA-accelerated market-data handlers and quoting engines.
• Integrated secure on-premise LLM tooling with Git/Jira/Confluence to automate code, testing, and documentation workflows.

• Architected Map Construction, Map Routing, and Rich Vehicle Routing algorithms (3 nested NP-Hard problems) using CP-SAT constraint programming and convex optimization over PostGIS, MongoDB, and S3.
• Led a 12-engineer team delivering a high-throughput geospatial mapping application platform.
• Built an LLM-powered debugging and query-resolution tool used company-wide, cutting mean bug-resolution time by 80%.

• Developed and backtested systematic merger-arbitrage strategies for an $8.5 Billion AUM fund, improving alpha capture by 15%.
• Built and deployed ML pipelines for Order and Execution Management Systems, increasing trade execution efficiency by 29%.
• Designed an ESG-driven merger-arbitrage signal capitalizing on pre- and post-merger statistics.

• Engineered Python-based trading services enhancing storage, processing, matching, and execution of trades on QUARTZ.
• Integrated C++ pipelines to store trades in the object-oriented database SANDRA, reducing trade processing latency by 50%.
• Led the migration of 1 million+ lines of code to Python 3.8, enhancing scalability and execution efficiency by 40%.

• Architected and developed an ML/AI platform to deploy predictive models, increasing decision-making accuracy by 67%.
• Designed machine learning models for data validation rules prediction, reducing workload by close to 36 Full-Time Equivalents (FTEs).

Sub-10us low-latency trading system with a custom-built limit order book, FPGA market data handlers, kernel bypass (DPDK), hardware timestamping, and lock-free data structures for deterministic, microsecond-level execution.

Local-first, autonomous agentic AI platform orchestrating LLM providers behind a unified API with MCP tool-calling, RAG and persistent semantic memory on ChromaDB, and hardware-aware deployment of quantized open-weight models.

Cross-platform agentic AI developer platform on NixOS with declarative configuration, multi-agent orchestration, isolated Git worktrees, autonomous task execution, and CI-gated shipping.

Adaptive volatility regime-switching framework selecting among passive, TWAP, and aggressive execution: +20.0% Sharpe Ratio, -6.1% transaction costs, -20.1% CVaR.

120-day volume-momentum crypto portfolio strategy: 155.76% annualized return and 1.94 Sharpe Ratio post transaction costs, outperforming the Bitcoin buy-and-hold benchmark.

Real-time portfolio optimization with convex and non-convex methods, adaptive rebalancing, and multi-factor modeling across interest rate, FX, credit, and market risks.