I'm Naveen Babu Narayanan Nair, based in Philadelphia. I build engineering teams, write papers, publish open tools and speak at conferences.
I lead the engineering teams behind device software platforms at Comcast.
Release cycles cut from six months to two weeks
ResearchI study how agentic AI can operate networks and run software tests, and I review papers for ACM.
Agentic Predictive Network Operations, Agentic Traffic
Open sourceI publish open tools for working with AI agents and for writing research papers.
Tracedown finds why an AI agent gave a wrong answerResearch Paper Toolkit covers IEEE, ACM, Springer, Elsevier and NeurIPS-style papersproctor is the exam proctor for AI coding agents
SpeakingI talk to the RDK community about test automation and shared testing.
Amsterdam 2024 speaker, Bonn 2023 panelist
Leadership
Comcast, Philadelphia. October 2024 to present.
I lead the engineering organization that builds the platforms and tools behind the RDK device software lifecycle, from release readiness and deployment to monitoring and field triage.
The work spans an agentic AI test automation platform, a global device farm with labs in the US, UK and India, and a multi-tenant device management platform for RDK operators.
Research
Each one starts from a practical problem and asks what changes when software agents take part in the loop. The latest preprint follows the three.
Agents that read network telemetry, forecast faults and degradations, and act, or recommend action, before customers notice. The open question is how much autonomy to grant while operators stay in control.
A car warns you the brake pads are thin weeks before they fail, and books the garage. The network does the same with its own signals.
Treats traffic generated by AI agents as its own class of network workload, and asks how it can be identified, measured and verified so that networks can plan for it and operators can trust it.
Delivery robots join a busy sidewalk. Planners need to know which travellers are robots, how many there are, and whether they follow the rules.
Learning-based analytics that recommend and prioritise tests and audit coverage and decisions, so very large test suites stay fast, explainable and trustworthy as the software changes.
A doctor with one hour decides which tests to order first based on the symptoms, and writes down why. Test suites face the same squeeze.
The paper asks whether an open-weight language model that breaks a rule passes the habit on to models made from it by fine-tuning, merging, quantizing or distilling. I tested two rules, claiming to be human and disclosing personal data to an unauthorized caller, on Qwen2.5-7B and Llama-3.1-8B.
Descendants kept the behavior, and models distilled from rule-breakers broke the rules far more often than those distilled from clean models. Exploratory weight-signature tests linked nearly all descendants back to their source, with one false link in 194 comparisons.
Read the preprint on ZenodoA photocopy of a page with a smudge carries the smudge, and so does every copy made from it. The paper checks whether copies of a rule-breaking model carry its bad habit, and whether you can trace a copy back to the original.
Open source
A Python SDK that traces AI agent runs, finds the step that led to a wrong answer and lets you replay the run with a fix.
View on GitHubSkills and prompts for drafting, checking and reviewing papers in IEEE, ACM, Springer, Elsevier and NeurIPS-style formats. It also includes a plain-prompt pack for researchers who use other AI assistants.
View on GitHubThe exam proctor for AI coding agents. It checks for cheating such as skipped failing tests or hard-coded outputs, and runs the code against hidden tests the agent never saw.
View on GitHubSpeaking
Presented AUTOMATICS, an open-source test automation framework and orchestration system adopted by Deutsche Telekom, Vodafone and Vantiva.
Community Test Collaboration: how RDK members can share testing effort and results.
I enjoy conferences, meetups and university talks. These are the topics I can bring.
Send an invitationEverything an organizer needs for a programme or introduction.
Naveen Babu Narayanan Nair is Senior Director of Software Development & Engineering at Comcast, where he leads engineering for RDK device software operations. He researches agentic AI for network operations and software test, builds open-source tools, and is an ACM Certified Reviewer. He has spoken at the RDK Technology Summit in Amsterdam and Bonn.
Naveen Babu Narayanan Nair is Senior Director of Software Development & Engineering at Comcast in Philadelphia, where he leads the engineering organization behind RDK device software operations, from release readiness and deployment to monitoring and field triage. Over more than twenty years he has built test automation and developer tooling for large device ecosystems, including AUTOMATICS, a test automation framework and orchestration system that was open-sourced to the RDK community and adopted by Deutsche Telekom, Vodafone and Vantiva. He researches agentic AI for network operations and software test, and he publishes open-source tools for researchers. Naveen presented at the RDK Technology Summit in Amsterdam in 2024 and joined a panel on community test collaboration in Bonn in 2023. He is an ACM Certified Reviewer, mentors early-stage founders with Kerala Startup Mission, and holds a Master of Computer Applications from Visvesvaraya Technological University.
About
For over twenty years I've built test automation and developer tooling for large device ecosystems. One question kept coming back: how can a complex system notice its own problems and explain them?
Today that question takes the form of agentic AI, applied to network operations and software test. I publish what I can, open-source what I can, and take the ideas to the RDK community, where operators share testing tools and results.
I'm an ACM Certified Reviewer and mentor early-stage founders with Kerala Startup Mission. I grew up in Kerala and now live in Philadelphia.
What I research
What I build with
How I work with people
Contact
For speaking invitations, research collaboration, mentoring questions or code reviews, write to me directly or use the form.
