Founding Quantitative AI Researcher · QuantLink AI
I build systems that turn messy market data into decision-grade alpha: factor engines, ML strategy research, derivatives analytics, and NLP signals, taken from first dataset to production interface.
$ whoami kanav_mehta · quantitative_researcher focus = factors | portfolio_construction | risk | agentic_research stack = python · sql · lightgbm · transformers · mcp mode = research → production, with audit trails
Kanav MehtaLos Angeles · UCLA MFE
00 · Flagship
Engine · Desk · Reports
A survivorship-bias-free equity engine, an eleven-specialist analyst desk that debates and sizes each case, and a report generator rendering chart-led outlooks, all in one gold/teal visual identity. Latest outlooks: MSFT, GOOGL, META, and NVDA 12-month flagships, each argued across 20+ exhibits.
01 · Current Role
2026–Present · Quant Analytics
At QuantLink AI I'm building an agent layer for quantitative research (an MCP-based platform where AI agents research, evaluate, and refine investment strategies with the tools of a quant desk) alongside constrained portfolio construction and multi-factor risk monitoring.
Engineering impact
GitHub · Mar–Nov 2026
Monthly contributions climbed from a few hundred over the summer to 2,120 in September and 2,726 in October, while the median pull request still went from open to merged in 5.4 hours.
02 · Focus Areas
Factors · Strategies · Language
QuantLink AI
Systematic Trading
AI + Markets
03 · Experience
2017–Present
QuantLink AI
Building an MCP-based agentic quant research platform (AI agents that research, evaluate, and refine strategies through desk-grade backtesting, portfolio construction & optimization, and risk-model attribution) alongside constrained long-only/long-short construction with multi-factor exposure monitoring. Shipped a point-in-time market-data platform, a SQL-native backtester, AutoML Workbench, Notebook Dashboards with a self-checking notebook AI agent, an options workspace, and a 1,139-operator typed catalog, with 517 merged pull requests and 1,059 code reviews to date.
Quanta Ventures
End-to-end Python ML pipeline for Bitcoin, Ethereum, and QQQ strategies: sliding-window LightGBM linear trees, walk-forward validation, live paper trading net of transaction costs with Sharpe above 1.5. Partnered with traders on equity-index, crude-oil, and precious-metals futures research.
WorldQuant
Engineered 40+ intraday and medium-frequency strategies from fundamental, analyst, short-interest, news, price-volume, and options data across US, China, Europe, and UK markets.
Wolfe Research
Built NLP factors, risk-neutralized models, intangible-asset research, macro risk-exposure frameworks, and analyst-facing dashboards for alpha discovery and portfolio context.
ZS Associates · Udaan
Observational studies on claims and EHR data, survival analysis, anomaly detection, revenue forecasting, and operational collections optimization. The statistical foundation for everything since.
04 · Selected Work
Full index → /projects
An agent desk that researches, debates, and sizes a single-name case, then prints it. Eleven specialists, a recorded bull/bear debate, and a risk-manager veto before any outlook ships.
Cross-sectional factor dashboard: ranking, diagnostics, buckets, risk, and backtest review in one workflow.
No-code training, ranking, explanation and deployment of many ML models, with leakage-safe purged walk-forward validation and holdout Rank IC leaderboards.
One click from Python research notebook to interactive dashboard: typed inputs, scheduled reruns, and server-side pivots over up to 50M rows.
Survivorship-free market data: a universe that keeps delisted companies, ~180M rows of calculations at exact parity, and a replayable dbt lakehouse.
Backtesting written entirely as dbt SQL on ClickHouse: no Python loop, look-ahead blocked by construction, P&L reconciled to every trade.
An AI agent that edits research notebooks against a live kernel and checks every edit before it applies, on an isolated, production-grade runtime.
FinBERT-FOMC suite released as a package and CLI: policy-regime tone, novelty scores, walk-forward validation, and sign-signal backtests.
End-to-end alpha engine for US equities on public data, designed to be driven by any MCP-capable AI agent: ingest, transform, backtest, optimize. Now at v0.5.0 with a survivorship-bias-free point-in-time universe.
RMT-denoised and shrinkage covariance estimation with transaction-cost-aware convex optimization for a copula-based US equity pairs strategy.
IAQF 2025 research: did SMB factor behavior decouple once concentration moved into mega-cap winners?
Long/short equity built on LEAPS instead of stock: Black-Scholes on escrowed spot, a calibrated IV proxy, and a daily-accounting engine that reprices exercise, expiry, and ruin on public data only.
Two market-implied AI return factors built point-in-time from ETF baskets: AI Theme (AUM-weighted, ex-semiconductors) and AI Infrastructure. Together they form the base for measuring firm-level AI exposure with Fama-French controls.
One optimization API across equities, crypto, futures, FX, and treasuries, unifying PyPortfolioOpt, skfolio, and Riskfolio-Lib behind native solvers: exact SOCP max-Sharpe under asset-class caps and turnover, CVaR LP, HRP/HERC, Black-Litterman, and a no-lookahead walk-forward backtester.
MCP-driven report engine that turns Markdown and chart PNGs into publication-grade PDF, HTML, and DOCX, with 19 templates, native Exhibit numbering, and mechanical quality gates that refuse to ship a bad exhibit. Renders QuantFlow's flagship outlooks.
Creativity and institutional-learning MCP server for quant research. Spark remembers everything a team has tried (failures included), deflates significance for the full trial count, and maps the unexplored corners of research space.
Fair-value engine for Kalshi's "Mentions" contracts: will a word be said, and how many times. A hurdle model pairs shrunk Beta-Binomial base rates with a truncated negative-binomial repetition process, plus live Gamma-Poisson updating mid-event and walk-forward backtests against the base rate.
05 · Research & Writing
Working out loud
Distribution, concentration, rank quality, bucket backtests: the checklist a factor must survive.
Top-bottom spreads can look clean until sector concentration turns a factor into an accidental macro bet.
Lattices, LSMC, finite differences, Monte Carlo: the method choice determines which risk surfaces you can see.
06 · Capabilities
Factor investing, cross-sectional analysis, derivatives pricing, fixed income, statistical arbitrage, portfolio construction, and risk diagnostics.
LightGBM, Elastic Net, transformer models, FinBERT, anomaly detection, and model validation under noisy market data, plus research agent workflows built on MCP tool servers.
Python, R, SQL, C++, Git, AWS, Linux; point-in-time data discipline, structured pipelines, APIs, and research reproducibility.
Interactive dashboards, chart design, stateful controls, screeners, backtest explorers, and analyst workflows built for repeated use.
07 · Education
UCLA Anderson School of Management
2024–2025 · GPA 3.93 / 4.0
IIT Roorkee
2013–2017 · KVPY Scholar
Contact
Los Angeles · UCLA Anderson MFE · research experience at QuantLink AI, Quanta Ventures, WorldQuant, and Wolfe Research.
