I design and build agentic AI platforms and network automation systems: the kind built to stay running, not just to demo well.
Hands-on videos from my YouTube channel. The player below always shows the latest uploads; use the playlist icon in the player to browse them all.
Practical sessions on network automation, SONiC and agentic AI for infrastructure, recorded so you can follow along at your own pace.
I'm a software engineer with 17+ years of hands-on experience designing and shipping production systems. Most recently that's agentic AI platforms, built during my current role at STORDIS GmbH; before that, network automation and management systems for telecom infrastructure, across STORDIS and earlier roles.
I work by one idea: a system is only as good as its worst day in production. That's why I give agents guardrails before I give them autonomy, and why I test automation tooling against failure, not just the happy path.
Five areas of hands-on depth, each built through production work rather than coursework or theory.
Designing and operating multi-agent AI platforms end to end: provider-agnostic LLM routing, retrieval engines built from scratch, and human-in-the-loop guardrails in front of any automated action.
Designing abstraction layers, APIs, and controllers that expose programmable network hardware to applications, including fault-tolerant SDN systems built for real traffic, not demos.
Delivering management platforms built to run reliably at scale, including systems rated for tens of thousands of managed devices, across distributed engineering teams.
Building monitoring and telemetry tooling, from protocol-level data collection across thousands of remote elements to AI-driven anomaly detection that flags drift before it becomes an outage.
Leading small, cross-site engineering teams through architecture, code review, and production on-call, staying hands-on with the code rather than delegating it away.
Built during my engineering roles, most recently at STORDIS GmbH; shown here as evidence of hands-on skill, not as independent commercial work.
Open-source network telemetry tooling: a custom metrics exporter streaming live switch state, and an agentless pipeline using AI-driven anomaly detection to flag configuration drift and predict hardware failure.
A multi-agent workflow system with provider-agnostic LLM routing, a dependency-free retrieval engine (TF-IDF and cosine similarity) built from scratch, and human-in-the-loop guardrails requiring confirmation before any state-changing action.
Built for the broadcasting industry's shift from SDI to IP networking: fault tolerance, traffic engineering, QoS and multicast forwarding on a P4-programmable data plane.
A notebook-based testing environment for validating network configuration protocols (RESTConf, OpenConfig) and prototyping routing behavior directly against live hardware.
More on github.com/kamalkrbh →
What I reach for day to day, grouped by where it sits in the system.
Conference talks at the Open Compute Project and CLOUDFEST, plus appearances as STORDIS's open networking and SONiC expert.
An AI-driven approach to designing, testing and automating network infrastructure with LLM agents and open networking (SONiC). Given as Head of Value Driven Software Delivery, STORDIS.
Monsoon 2.0 (agentless gNMI telemetry with AI-driven anomaly detection), AI-enhanced ORCA orchestration, and Smart CLI, an LLM assistant for the SONiC NOS.
An interview on open networking, featured on the STORDIS YouTube channel as the company’s open networking and SONiC expert.
How open-source projects SONiC, Monsoon (Prometheus-based SONiC monitoring) and AI-enhanced ORCA (orchestration with a LangChain assistant) drive innovation in open networking.
Monitoring and telemetry for open networking switches running SONiC, representing STORDIS.
More on the STORDIS channel and my own channel →