🌐 Website: dhyeymavani.com
📫 Email: ddmavani2003@gmail.com
💼 LinkedIn: linkedin.com/in/dhyey-mavani

• Agentic AI MLE (Special Projects) @ LinkedIn

  • Building the next generation of digital work AI agents directly with LinkedIn's Chief AI Officer, President of Engineering, and CEO. Youngest MLE on the team, working daily alongside Distinguished Engineers. Dabbling across product, GTM, and governance — context, tools, and agent infra.
  • Founded and led a pod of 3; co-authored research on LLM robustness (coined "Support Tokens," stability margins) under review at COLM 2026. Promoted to Highly Confidential capacity with access to broader MSFT + OpenAI resources.

• Writing — strategy and technical pieces published at VentureBeat, DZone, LeadDev, and Hacker News; Featured in Business Insider (Nov & Dec 2025) and other major industry media.

• Previously SWE – GenAI Inference @ LinkedIn
  Shipped n-gram speculative decoding into production for Hiring Assistant: 4× throughput, 66% latency reduction. Built LinkedIn's first MCP integrations with Azure OpenAI infra. Built fine-grained observability across the vLLM + SGLang hybrid serving stack with a centralized dashboard. Recruited and mentored Masters & PhDs on LLM serving.

• Previously SWE – GPU RAR (RecSys) @ LinkedIn
  Pitched 3 novel product ideas directly to CEO Ryan Roslansky; led to further discussions with product leaders across 3 business units on future B2B2C strategy. Built end-to-end Rust-based inference engine for PyTorch EBR kNN; reduced P99 latency from 4ms → 2ms, tail ingestion from 70ms → 7ms, ~$1M GPU savings serving 5+ use-cases.

• Applied Researcher @ Amherst College (4+ years)

  • chipfiring — first Python package for chip-firing on multigraphs. Featured in SIAM News.
  • ccrvam — discrete copula method for multivariate EDA. 30k+ downloads. Breusch Prize.
  • quantileflow — quantile sketching for service logs, 44% more efficient than Datadog.
  • chip-firing-with-lean — first-ever Lean4 formalization of chip-firing dynamics and Graphical Riemann–Roch.

• Y Combinator AI Startup School Fellow (2025, 2026) — invite-only, sponsored.

• Triple Major @ Amherst College in Computer Science, Mathematics, and Statistics (2× Summa Cum Laude, Phi Beta Kappa, Sigma Xi, 3.97 GPA).

• Honors Theses

  • 🧮 Mathematics: Lean4 formalization of chip-firing games & Graphical Riemann–Roch (believed first-ever formalization; preprint forthcoming)
  • 📊 Statistics: ccrvam – model-free EDA for multivariate discrete data. (Robert H. Breusch Prize for Best Honors Thesis in Statistics.)

• Previously Engineering @ AWS CloudFront, Quantitative Trading @ Valkyrie Trading. Former President of Amherst Quant Club & Founding Director @ i2i student venture studio.

🏸 Badminton — Yonex Northeastern Collegiate (6th in singles, wins vs Harvard/Binghamton)
🏹 Archery
🏏 Cricket (founded Amherst club)
🌊 Jetskiing, ATV riding, bungee jumping, zip-lining

Previous affiliations: @aws @minerva-ed @logflow-ai @AmherstCollege @acdmammoths @r-blocks @ac-i2i-engineering @ACQuantClub