Krish
Parmar.

Engineering at Cumulus Labs

YC W26
Carnegie Mellon UniversityStats & ML ’30
View WorkGet in Touch

Work History

Experience

Cumulus Labs

YC W26

Engineering

Nov 2025 — Present

San Francisco, CA

  • Co-engineered a GPU inference optimization layer achieving a 4.2× reduction in cold-start latency compared to industry baselines, significantly improving time-to-first-token for large models.
  • Designed and built a Kubernetes-native GPU scheduling and model-serving platform supporting fractional GPU sharing.
  • Implemented an intelligent job scheduler with batching, priority handling, and contention detection, reliably coordinating tens of concurrent inference and training workloads per cluster.
  • Built automated job monitoring and recovery logic that detects failed GPU workloads and dynamically adjusts execution parameters and resource assignments.
  • Delivered end-to-end production infrastructure spanning gRPC APIs, Kubernetes, per-node GPU agents, and a real-time admin console for monitoring jobs and GPU utilization.

Petra Security

YC W24

Software Engineer Intern

Jun 2025 — Aug 2025

San Francisco, CA

  • Engineered a production-grade phishing-detection system, implementing advanced string-matching algorithms and multithreaded processing to automatically identify and remove malicious emails, preventing recurrence of attacks.
  • Refactored backend services and optimized SQL queries, leveraging indexing strategies and query plan analysis to reduce API response times by 15–20% across high-volume endpoints.
  • Automated cloud-based trial environment provisioning using custom scripts and API integrations, integrating CI/CD workflows to cut enterprise onboarding time from days to hours.

Code Wiz

Software Instructor

Jun 2023 — Sept 2023

Westford, MA

  • Instructed 100+ students in programming, algorithms, and robotics, adapting lessons to varying skill levels.
  • Designed hands-on coding and robotics projects reinforcing key concepts through project-based learning.

What I've Built

Projects

01
2nd Place — YC Hackathon

VideoEditor MCP

AI Agent · Video Editing

An MCP App where an AI agent edits video inside DaVinci Resolve using natural language. Tell it to make a highlight reel and it does it — powered by Cumulus Labs’ Qwen3-VL-32B vision model to understand video frame-by-frame, and a Python bridge for frame-precise DaVinci edits. Won 2nd place at the Manufact (YC S25) MCP Apps Hackathon at Y Combinator.

View on GitHub→
02

Nova

Cofounder · DECA Prep Platform

Co-founded and launched a web app for DECA competition preparation. Cloud-hosted database managing 8,000+ questions with user auth, caching, and analytics. Live at novadeca.com.

360+

Active Users

1K+

Weekly Visits

8K+

Questions

Visit novadeca.com→
03

FRC Robotics

Team 5422 Stormgears

Architected Java-based swerve drivetrain and autonomous path-planning software. Integrated Limelight + Google Coral TPU computer vision for real-time object tracking and autonomous note acquisition.

#1

NE UNH Rank

30-17

Match Record

Top 15%

Worldwide EPA

View on GitHub→

Tools of the Trade

Stack

Languages & Libraries

PythonTypeScriptJavaGoSQLPostgreSQLNumPypandas

Tools & Frameworks

KubernetesDockerReactNext.jsGraphQLPrismaFlaskTailwind CSSgRPCPostman

Recognition

Achievements

EmailLinkedInGitHub

© Krish Parmar 2026