22+ years building telecom and AI systems — from kernel-level packet processing to agentic AI platforms. Architect who has taken 3 products from R&D to commercial deployment, including the world's first commercial 3GPP Femto Gateway, the world's first 5G-NSA deployment at T-Mobile, and supporting Reliance Jio's greenfield all-IP LTE network.
Unique career arc spanning every layer of the technology stack: bare-metal C/C++ with DPDK, cloud-native Kubernetes microservices in Golang, and now full-stack AI/LLM/Agentic infrastructure. Expert at adopting open-source into enterprise solutions — accelerating development and cost optimization. Pioneered CI/CD, observability platforms, and test automation frameworks before industry standardization (pre-Jenkins, pre-Prometheus era).
Currently driving strategic AI adoption at Ribbon Communications: built on-premises LLM clusters (12 GPUs, 230GB VRAM), deployed RAG chatbots saving $100K+/year, engineered agentic RFx systems reducing sales response from months to days, and championing agentic software dark factories — where 16+ years of architecture experience — writing PRDs, FSDs, DSDs, and revision design specifications — now directly fuel autonomous code generation and software development at 100x velocity.
From user-space AI applications down to bare-metal hardware — 22 years of expertise at every layer
16+ years of writing 3GPP-compliant specs means I produce specifications with the precision that AI agents need. Ambiguity kills agent performance — my specs eliminate it.
From DPDK kernel-level code to Kubernetes orchestration to LLM fine-tuning — I understand every layer an agent might touch.
Built Acumen AI Platform (FastAPI + Celery + MCP Servers) as the backbone for agentic workflows. Deployed Langflow for low-code agent orchestration.
Fine-tuned LLMs using LoRA/QLoRA/RLHF on product-specific knowledge. Built RAG systems over 100GB+ of domain documentation.
Built and operated a 12-GPU, 230GB VRAM AI cluster. Introduced agentic coding tools across multiple teams.
Built CI/CD before Jenkins. Built observability before Prometheus. Now building agentic software factories before they're mainstream.
A software dark factory is a development pipeline where AI agents autonomously generate, test, deploy, and monitor code. My career has been a perfect preparation for this: 16+ years of writing machine-readable specifications, building CI/CD and observability infrastructure, and now deploying the AI/LLM infrastructure that powers the agents themselves. The result: 100x the output of traditional development.
Architected analytics pipeline processing 3B events/hour, serving 2.5M customers at Airtel India. Full 5G stack support. Designed to scale to 400B events/hour.
3B events/hour → 400B target5-node K8s cluster with 12 GPUs, 230GB VRAM, 192 virtual GPU slices. Self-service AI platform — a scaled-down Amazon Bedrock equivalent.
12 GPUs | 230GB VRAM | 192 vGPUOn-premises LLM deployment. 100+ GB knowledge base, 10M+ tokens served, $100K+ annual savings with full data sovereignty.
$100K+ annual savings | 10M+ tokensAgentic AI handling 15,000 technical questions in minutes. Reduced sales first response from months to days. Accuracy: 75% to 87%.
Months → Days | 75% → 87%Process Terabits/second using Intel Tofino switch with P4 programming. Metadata creation for ML-based network analysis.
Terabits/second processingArchitected 5G-NSA solution for world's first 5G-NSA deployment at T-Mobile.
World's First 5G-NSA DeploymentCisco's industry-first Cloud Native Standalone 5G Core. Owned QoS Modeling (heart of SMF).
Industry-First Cloud-Native 5G CoreWorld's first commercial 3GPP Femto Gateway for SFR, France. 60,000 APs with 99.999% availability.
World's First | 60K APs | 99.999%TAC lead for world's first all-IP 4G/LTE network — 125K gNodeBs. KPIs for 100,000+ eNodeBs.
World's Largest All-IP LTE NetworkSemifinalist among 400+ submissions (top 20) in Cisco CTO organization.
Top 20 of 400+ submissionsBeyond hands-on engineering, I serve as a trusted AI strategy advisor to senior leadership — helping VPs, SVPs, Product Managers, and CXOs gain clarity on AI strategy, investment decisions, and technology roadmaps.
Delivered customer-facing POC demos at Mobile World Congress — AI-powered analytics, agentic AI platforms, and network intelligence showcases.
Consulted by CXOs, SVPs, and VPs to translate complex GenAI/LLM concepts into actionable business strategies. Build vs. buy decisions, investment prioritization.
Prepared AI strategy presentations, technology roadmap slides, and competitive analysis decks for senior leadership.
Core member driving enterprise GenAI adoption. Introduced developer tools (Codeium, Windsurf, Cursor, Copilot), moderated internal AI/ML forum.
Guided PMs on integrating AI capabilities into product roadmaps — identifying features that benefit from LLMs, RAG, fine-tuning, or agentic workflows.
Conducted product viability assessments, MVP definition, and competitive intelligence for next-gen analytics and AI products.
The ability to bridge deep technical expertise with strategic business communication differentiates an architect from a technology leader. From MWC demos that win customer interest to slides that give CXOs clarity on multi-million dollar AI investments — this strategic advisory role amplifies the impact of every technical capability.
Active personal AI experimentation lab — a living cookbook of hands-on projects exploring cutting-edge AI/ML techniques. Each experiment is designed to deeply understand underlying mechanisms.
Real-time voice agent using LiveKit WebRTC, Sarvam STT/TTS, and local Gemma3:4B via Ollama. Complete pipeline: streaming audio → STT → LLM reasoning → TTS → voice response.
LoRA and QLoRA on Llama 3.2 3B with 4-bit quantization, PDF-to-QA data pipelines, GGUF export for Ollama, ROUGE/BERTScore evaluation.
RAG-powered compliance system using DeepSeek-R1:8B, LangChain, Chroma. Multi-stage retrieval: cosine similarity → cross-encoder re-ranking → fuzzy matching.
Built custom Langflow components: KB List Fetcher, Ollama integration, regex processors, URL processors, and OpenAI wrappers.
Golang-to-Celery bridge — Go as high-performance backend submitting async tasks to Python Celery workers via Redis.
Role-based access control using Firebase Authentication with FastAPI — JWT verification, dependency injection, user DB management.
Full NLP/ML pipeline: Text Generation with TensorFlow/Keras/LSTM, summarization, POS tagging, classification, Text-to-Image with HuggingFace.
Intelligent Excel-to-vector-DB ingestion: logical table detection, structure-preserving markdown conversion, smart chunking respecting table boundaries.
Optimized Quality of Service Enforcement for Handover Between Different Radio Access Types
US Patent — Granted.
LangChain Academy | December 2025
LangChain Academy | December 2025
DeepLearning.AI | October 2025
DeepLearning.AI | September 2025
Udemy | February 2025
Udemy | February 2023
Date of Birth: March 28, 1983
Languages: Kannada, English, Hindi
Interests: Technology exploration, photography, traveling, reading, music
Teaching and enabling others has been a steady part of the architect role across Nokia, Samsung, Cisco, and Ribbon — onboarding new engineers, running company-wide technical training, guiding university students through AI capstone projects, and helping engineering and sales teams reach working proficiency on new tools and agentic workflows, not just theoretical familiarity.
Ongoing responsibility across every role — ramping new team members on codebase, architecture, 3GPP protocols, and debugging workflows through hands-on pairing until they are independently productive.
Mentored graduate, post-graduate, and PhD students on AI capstone projects — scoping problem statements, guiding LLM/RAG/agent design choices, evaluation methodology, and reviewing final deliverables.
Helped engineering teams adopt agentic flows for log analysis and reasoning agents — walking through prompt design, tool wiring, memory and context patterns, and observability so teams can extend the agents themselves.
Delivered in-person technical training sessions to audiences of up to 50 attendees — covering telecom architecture topics, AI/LLM fundamentals, and new platform capabilities.
Company-wide responsibility for assessing participant readiness and training teams on on-prem AI infrastructure ahead of internal AI hackathons — cluster access, LLM and framework selection, agent tooling — so teams spend hackathon hours on ideas, not setup.
Partnered with program management and sales to scale up adoption of AI tools — translating technical detail into what engineering needs to execute, and distilling solution options into what sales needs to position confidently.
The move to agile and a handcrafted CI/CD toolchain was driven through hands-on training — building the tooling alongside the 120-member team so the new way of working became a working skill, not a slideware change.
The distinction that runs through all of these engagements: the goal is always working proficiency, not theoretical familiarity. Whether onboarding a new graduate engineer at Nokia, training a sales team on AI positioning, or guiding a PhD student through RAG architecture choices — the measure of success is whether the person can operate independently afterward. The architect role is a force multiplier; so is teaching.