Applied AI Engineer | ML Platform | Agentic Workflows | LLM Pretraining, Fine-Tuning, RLHF | MLOps

I build production-grade ML and AI systems across workflow orchestration, RAG and agent runtimes, LLM pretraining and post-training, release governance, and Kubernetes-native platform operations.

My work is centered on:

• Agentic workflow platforms with typed execution contracts, tools, capabilities, memory, and triggers
• LLM and RAG systems with evaluation, observability, and operational guardrails
• LLM training and adaptation workflows including pretraining, supervised fine-tuning, and RLHF-style preference optimization
• ML platform and release pipelines with measurable quality gates and controlled promotion
• Local-first developer workflows that map cleanly to cloud and Kubernetes production patterns

All public repositories use educational, synthetic, or non-sensitive data only.

• AI workflow platforms and orchestration runtimes
• Tool-using agent and retrieval-augmented systems
• LLM pretraining, fine-tuning, and RLHF-style alignment workflows
• ML and LLM release governance pipelines
• Observability-first platform services

A full-stack platform for authoring and running AI-powered workflows through chat and a visual DAG editor, backed by typed execution contracts, reusable capabilities, memory, triggers, and Kubernetes-native orchestration.

Highlights

• Chat, Compose, and Workflow Studio surfaces for conversational, goal-driven, and manually authored workflows
• Typed planner/worker runtime with reusable capabilities, memory integration, control-flow nodes, retries, and DLQ recovery
• Kubernetes-ready scaling, artifact/document handling, and observability with Prometheus, Grafana, Loki, and Jaeger

Repo: https://github.com/narendersurabhi/agentic-workflow-studio

A Kubernetes-first release governance project focused on controlled promotion, plan/execute workflows, policy gating, and observable ML operations.

Highlights

• Release-gate workflow for evaluation, approval, and promotion
• MLflow-backed artifacts and runtime decisioning
• Prometheus and OpenTelemetry instrumentation with Jaeger tracing

Repo: https://github.com/narendersurabhi/mlops-release-gate-agent

A production-style MCP server with multiple transports, auth policy controls, and observability built in from the start.

Highlights

• Stdio and HTTP MCP transport support
• Scope-based bearer-token authorization model
• OpenTelemetry and Prometheus instrumentation for runtime visibility

Repo: https://github.com/narendersurabhi/mcp-control-plane

An end-to-end repository for LLM customization workflows spanning fine-tuning, RLHF-style preference optimization, evaluation, and serving.

Highlights

• LoRA and QLoRA supervised fine-tuning flows
• DPO-style preference optimization and evaluation gates
• Production-facing API and observability patterns for model serving

Repo: https://github.com/narendersurabhi/llm-customization-ops

A reference implementation for model governance and promotion across training, evaluation, registry flow, and serving.

Highlights

• Train, evaluate, promote, and serve workflow
• Artifact and model-registry oriented release path
• FastAPI serving with Prometheus and Grafana-compatible metrics

Repo: https://github.com/narendersurabhi/ml-platform-release-gates

• Contracts-first design for predictable component boundaries
• LLM lifecycle coverage from training and post-training through evaluation and serving
• Evidence-based release decisions using measurable criteria
• Operability as a hard requirement, not a follow-on task
• Reproducible local and CI workflows with deterministic test paths

• Platform and API: Python, FastAPI, Docker, Kubernetes, Helm, MLflow
• Observability: OpenTelemetry, Prometheus, Grafana, Jaeger, Loki
• LLM systems: pretraining concepts, supervised fine-tuning, RLHF and preference optimization, RAG, tool-calling agents, workflow orchestration, evaluation pipelines
• ML and data: PyTorch, XGBoost, PySpark, classical ML pipelines

• Applied AI Engineer
• AI Platform Engineer
• ML Platform Engineer

• LinkedIn: https://www.linkedin.com/in/narendersurabhi
• GitHub: https://github.com/narendersurabhi
• Location: Okemos, MI