Ajeet
Kumar
Engineering @ Google — agentic AI & ML systems, after 11+ years shipping the 5G/LTE stacks that carry them.
About me
I build systems that sit close to the metal and close to the deadline. Eleven and a half years at Qualcomm went into 5G/LTE protocol stacks — RLC, PDCP, NAS, MAC — the low-level plumbing that keeps a phone connected. I later pulled that same discipline for correctness under constraints into machine learning, shipping one of Qualcomm's first production edge-ML use cases directly on the modem.
At Google, I'm now co-engineering an agent framework from 0→1 in C++, powering reasoning loops for LLM-based systems, and running the TPU / large-model serving infrastructure behind Nest camera perception and real-time captioning.
Along the way: Global eCall Developer Lead across worldwide sites, a from-scratch rewrite of a multi-SIM codebase from three threads down to one, and five granted US patents in wireless and thermal-mitigation technology.
Where I've signaled from
- Co-engineering an agent framework from 0→1 in C++ to power complex LLM reasoning loops.
- Building Cloud ML services and managing TPU / large-model serving infrastructure for Nest camera perception and real-time captioning.
- Developed 5G spec features (ENDC/NRDC), PDCP UL Status PDU handling, and ROHC on the 5G L2 data path.
- Delivered Qualcomm's first production edge-ML use case on wireless modems — C++ modules integrated with AWS SageMaker.
- Filed 4 IDFs; built process-improvement tooling that removed manual intervention from the team's workflow.
- Delivered the NextGen eCall feature and simultaneous voice-and-data capability.
- Designed multi-SIM features from the Non-Access Stratum layer.
- Led a reorganization of the codebase from a three-thread to a single-thread architecture for low mem/CPU targets.
- Dev Lead for the eCall feature across Qualcomm's global sites.
- Designed and delivered multi-SIM features from the Non-Access Stratum layer.
- Key contributor on the Triple SIM project and SGLTE feature.
- Worked on multi-protocol system selection algorithms.
- Built LTE eNodeB MAC schedulers (QoS, round-robin) and Inter-Cell Interference Coordination features.
- Replaced a mesh Diameter Signaling Router topology with a single central router — UI and backend.
Notable work
Agent framework, 0→1
Co-built a C++ agent framework at Google powering complex, multi-step reasoning loops for LLM-based systems.
Click for details →Nest ML serving infrastructure
Owns TPU and large-model serving infra behind Nest camera perception and real-time captioning.
Click for details →First edge-ML on a modem
Shipped Qualcomm's first production ML use case running directly on wireless modem hardware.
Click for details →5 granted US patents
Wireless multi-SIM registration optimization and thermal-mitigation systems for multi-subscription devices.
Click to see all 5 →What people say
Placeholder — add a recommendation from a manager, teammate, or collaborator here. LinkedIn recommendations copy in well.
Placeholder — a second recommendation, ideally from someone at a different company to show range.
Placeholder — a third one rounds it out nicely. Keep each under ~40 words so the cards stay balanced.
These are placeholders — swap in real quotes (e.g. from LinkedIn recommendations) before publishing.
Skills & stack
Certifications & education
- Master of Computer Applications (MCA), Birla Institute of Technology — 2007–2010
- Generative AI with Large Language Models — DeepLearning.AI
- Structuring Machine Learning Projects
- Machine Learning A-Z™: Hands-On Python & R in Data Science
- AWS Machine Learning Engineer Nanodegree
Open to conversations on agentic AI, ML infra, or anything wireless.