MAYUR●
WORLD 1-1COINS 00LEVEL 01
WORLD 1-1
Pixel-art avatar of Mayur Kishore Kumar

Mayur
Kishore Kumar

MACHINE LEARNING ENGINEER

APPLIED AI/ML · GENAI · RAG · AGENTS · MCP

★ CURRENT QUEST

Find the next hard problem worth building.

GITHUBLINKEDINRESUME
SCROLL TO MOVE THROUGH THE WORLD · ↑ ↓ TO SKIP LEVELS
WORLD 1-1

ABOUT
THE PLAYER

QUEST: FIND OUT WHAT I BUILD

I’m a Machine Learning Engineer building production ML and AI systems across semiconductor and healthcare domains.

My work spans predictive modeling, RAG, multi-agent systems, MCP integrations, and ML infrastructure.

At Rambus, I build enterprise AI solutions for semiconductor engineering workflows in an air-gapped environment.

I enjoy taking ambiguous engineering problems and turning them into systems that people can actually use.

WORLD 1-2

POWER-UPS

COLLECT THE STACK

PY

MACHINE LEARNING

PythonSQLPyTorchTensorFlowScikit-LearnNumPyPandasXGBoost
AI

GENERATIVE AI

RAGAI AgentsMCPLangChainLangGraphEmbeddingsFAISSChromaDBPrompt EngineeringOllamaGroqn8n
⚙

ML INFRASTRUCTURE

LinuxDockerKubernetesMLflowFastAPIFlaskPostgreSQLAWSAzureGCPGit
⌘

COMPUTER VISION

YOLOOpenCVImage ProcessingPaddleOCREasyOCRTesseract
$
WORLD 1-3

LEVELS CLEARED

CHECKPOINTS FROM THE RUN

1.5M+ JOBSR² 0.8715 MCPs70 AI SKILLS
01
JUL 2025 — PRESENT · BENGALURU

Machine Learning Engineer Apprentice · Rambus

Production ML and AI systems across semiconductor engineering workflows.

  • Scaled EDA resource prediction from 30K → 1.5M+ production jobs across 4+ years; R² 0.87 for 90th-percentile resource prediction using quantile regression.
  • Designed enterprise RAG combining BM25 keyword retrieval, vector search, reranking and knowledge-graph retrieval, with feedback indicating 40% average time savings.
  • Built 15 MCP integrations personally (~20 across the team) for Jira, Confluence, Jenkins and ServiceNow, serving hundreds of users and supporting ~30% of relevant ticket activity.
  • Engineered AI developer tooling across GitHub Copilot, Continue.dev, terminal agents and OpenClaw, including 70 engineering skills and integrations supporting 80–100 daily users and millions of tokens/day.
  • Automated NetApp filesystem analytics, VS Code extension provisioning, SFTP semiconductor data pipelines, and GenAI log/document workflows used by 20–30 finance users.
02
SEP 2024 — MAY 2025 · BOSTON

AI Engineer · Humanitarians AI

Created 10+ AI applications from scratch using custom RAG and multi-agent systems.

  • Built clinical triage and patient-routing RAG pipelines across 10K+ documents, from ingestion and semantic search through reranking and prompt orchestration.
  • Engineered multi-agent decision-support workflows combining planning, memory and tool orchestration.
  • Benchmarked language models, embeddings and FAISS/ChromaDB for retrieval quality, inference cost and deployment performance.
  • Enabled local Ollama and cloud Groq inference across varying connectivity and compute constraints.
03
MAR 2025 — MAY 2025 · SAN FRANCISCO

Python Engineering Expert · AfterQuery Experts

Improved reliability across large-scale public Python codebases.

  • Diagnosed and resolved open issues across multiple GitHub repositories, submitting production-ready patches.
  • Replicated environments with Docker and developed 5+ custom test cases per fix to validate solutions and robust error handling.
04
MAR 2021 — JUN 2021 · BENGALURU

Project Trainee · Indian Space Research Organization

Built predictive monitoring models using 10+ years of telemetry data, enabling proactive overheating-risk detection and reducing response time by 30%.

WORLD 1-4

PROJECT CASTLE

CHOOSE YOUR BUILD

LEVEL 01 · OPEN SOURCE

MULTI-AGENT
AI REASONING

Extensible multi-agent reasoning framework integrating SQL databases, REST APIs and document retrieval using Groq and Ollama.

VIEW CODE →
LEVEL 02 · WORK PROJECT

EDA RESOURCE
PREDICTION

Production ML system using IBM LSF and Spectre data, scaled to 1.5M+ jobs with R² 0.87 for 90th-percentile resource prediction.

R² 0.87
LEVEL 03 · ML + XAI

EXPLAINABLE NBA
PREDICTIONS

XGBoost game outcome prediction with BERT, Llama and Mistral for natural-language analysis, plus SHAP and LIME interpretability.

85% ACCURACY
LEVEL 04 · DEEP LEARNING

LUNG CANCER
IMAGE CLASSIFICATION

CNN-based image classification pipeline on the LC25000 lung and colon histopathological image dataset.

VIEW CODE →
⚑
LEVEL COMPLETE

Ready for the next quest?

CONTACT MAYUR →
+1 COIN!