👨‍💻 About Me

🏎️ View My Tech Stack 🤖

🐋 Contribution Tulkun

_{🐋 Grows with my yearly contributions · regenerated daily by GitHub Actions}

📡 Engineering Telemetry

🏆 LeetCode Info

  

🧊 Capability Stack

🌌 3D Contribution Graph

_{✨ Contribution history rendered as an animated 3D constellation}

_{🟩 Name rendered in GitHub contribution-style green pixel blocks}

🏭 Production Engineering Focus

_{👇 Expand any domain — scope of ownership and how I measure it}

🔭 LLMOps Operating LLM systems like production software What I own Prompt, model and config versioning with controlled rollout and rollback End-to-end tracing across every model and tool call Per-request cost, latency and quality budgets Guardrails, fallbacks and routing for graceful degradation Measured by  ·  p95 latency · cost per task · regression rate	⚙️ MLOps Reproducible pipelines, automated delivery What I own Experiment tracking, data and model versioning, model registry CI/CD for training and serving pipelines Containerized, reproducible environments Drift and degradation monitoring with alerting Measured by  ·  reproducibility · time-to-deploy · time-to-detect drift
🧪 LLM Evals Quality measured before users find the gaps What I own Golden datasets and regression suites as release gates Calibrated LLM-as-judge rubrics backed by human review RAG metrics: faithfulness, context precision and recall Agent trajectory and tool-use evaluation Measured by  ·  gate pass-rate · judge–human agreement · escaped defects	🕸️ Multi-Agent Harness Engineering Orchestration built to survive production What I own Stateful multi-agent graphs, MCP servers and strict tool contracts Retries, timeouts and step budgets with human-in-the-loop checkpoints Sandboxed execution for untrusted tool calls A reproducible harness for evaluating agent behavior Measured by  ·  task success rate · steps per task · recovery rate
🧠 Transformers & Model Adaptation Understanding the model, not just calling it What I own Attention mechanics, tokenization and context-window trade-offs Parameter-efficient fine-tuning and domain adaptation Quantization, batching and KV-cache-aware inference Model selection driven by task-level evals, not leaderboards Measured by  ·  quality delta vs. baseline · tokens/sec · memory footprint	☁️ AWS & Cloud Infrastructure Deploy, secure and operate at scale What I own Serverless and container compute, managed data stores IAM least-privilege, secrets management and audit trails Infrastructure as code with automated, reviewable deploys Monitoring, alarms and cost visibility from day one Measured by  ·  deploy frequency · availability · cost per request
🔎 RAG & Retrieval Systems Grounded, measurable answers What I own Hybrid retrieval with reranking Chunking and indexing strategy tuned against eval results Production vector stores with freshness and access control Measured by  ·  faithfulness · retrieval recall · answer latency	🚦 Serving, APIs & Reliability The layer users actually touch What I own Async, streaming APIs with backpressure Queues, rate limiting and graceful degradation Prompt-injection defense and input/output validation SLO-driven alerting and incident readiness Measured by  ·  availability · error budget · p95 latency

🚀 Founder & Lead Developer — OpenML

Role	Founder & Lead Developer
Product	AutoML pipeline tooling — data in, trained and evaluated model out, from the command line
Install	pip install open-mlpipe
Documentation	openml-docs.vercel.app
Company page	OpenML on LinkedIn

🌍 Let's Connect

📍 Bhopal, Madhya Pradesh, India  |  🎓 AIML @ OIST  |  📧 [email protected]

⚡ Building Intelligent AI Systems Today — For a Smarter Tomorrow ⚡

_{Projects, live demos, and interactive 3D work — all in one place.}