From quantitative models to product roadmaps — I turn market complexity into systems that ship.
Unified fragmented collateral exposure data across global vendors into a single source of truth — eliminating ~80% of manual reporting, enabling tracking of $500M+ in daily exposure, and achieving global adoption across risk, legal, and trading teams.
Replaced fragmented spreadsheets with real-time PnL visibility across 10+ global trading desks. Cut reconciliation time by ~75%.
Unified fragmented on-prem and cloud ML workflows into a Kubernetes-orchestrated platform with standardized CI/CD pipelines.
Policy-driven system modeling capital deployment and collateral monitoring for lender-funded investment platforms, with a simulation layer for scenario testing.
Human-in-the-loop evaluation framework for AI agents at Meta — multi-dimensional scoring and feedback loops that improved response quality by ~25%.
Embedded LLM-powered anomaly detection into compliance workflows — cutting audit cycles by 40% and manual review work by 60%.
Automated credit and margin workflows across 33 counterparties — saving $2M+ annually and eliminating 99.5% of manual effort.
Deployed NLP-powered semantic search across a risk analytics platform for 5,000+ users — boosting engagement by 25%.
Unified Bloomberg, Yahoo Finance, and Yieldbook ingestion into a canonical data layer — cutting latency 35% and enabling real-time trade visibility.
Early-stage ideas explored through design thinking — problems worth solving, decisions worth making.
Tokenizing real-world bonds into fractional, programmable financial assets — bringing institutional-grade fixed income to retail investors via ERC-3643 compliance standards.
Read writeup ↗A centralized financial health record that aggregates client data, documents, and AI-generated insights into a single advisor interface — built on GraphRAG.
Read writeup ↗A multi-agent system for LPs and GPs that continuously monitors portfolios and surfaces actionable nudges for allocation, risk, and liquidity — moving beyond static dashboards.
Read writeup ↗Mental models from the quant desk, applied to product decisions.
Every feature decision is a trade-off. When prioritizing a new algorithmic order type, the real question is: what are you not doing? Deferring a stability fix in the clearing system isn't free — it carries a hidden cost measured in downtime risk, trader trust, and potential P&L impact.
In trading, gut instinct gets quantified. Backtesting and simulated environments let you stress-test a hypothesis before committing capital — or engineering time. The same discipline applies to product: run the numbers, simulate the scenario, and let P&L impact inform your roadmap before you build.
"Numbers tell you what happened.
Products shape what happens next."
I started my career as a Quant Engineer — building models, backtesting strategies, and turning data into decisions. That foundation gave me a rare lens when I moved into product management: I think in systems, challenge assumptions with data, and stay relentlessly focused on measurable outcomes.
Today I work at the intersection of technology and strategy, helping teams build products that are both analytically sound and deeply human. When I'm not shipping product, I'm curating the next wave of technical content for my Substack video catalog.
Whether you want to collaborate, discuss products and data, or just say hello — my inbox is open.
hello@smithashivakumar.com