Senior AI Consultant & Architect

Where 5G Systems Meet
Generative AI

9+ years building production-grade AI platforms at the intersection of telecom engineering and LLM architecture. I design systems that make complex specifications, logs, and workflows intelligently accessible — at scale.

Agentic AI RAG Pipelines LLM Orchestration 5G NR / 3GPP AWS Bedrock LangGraph Healthcare AI 3 Patents
9+
Years experience
3
Patents filed
30s
3GPP query time (vs 45 min manual)
5G+AI
Unique domain combination
Key projects
All architectures are conceptual representations of real production systems. No proprietary code or data is shared.
RAG Knowledge Systems Telecom
Hybrid RAG Platform for 3GPP & Telecom Knowledge Systems

A telecom-focused retrieval-augmented generation platform purpose-built for reasoning over 3GPP specifications (TS 38.xxx, 23.xxx, 29.xxx). Combines semantic search, keyword search, and metadata filtering with graph-based relationship modelling for high-precision, auditable AI responses.

Architecture overview
User Query Natural language 3GPP question Hybrid Retriever Semantic search Keyword + metadata Knowledge Store S3 Vector Store Amazon Neptune 3GPP TS/TR corpus Delta chunking LLM Grounding AWS Bedrock Claude / Mistral Auditable Response Evidence-grounded
Outcome
Reduced 3GPP specification query resolution from ~45 minutes of manual search to under 30 seconds, deployed across the engineering team.
AWS BedrockClaudeS3 Vector StoreAmazon NeptuneHybrid SearchPythonFastAPI
Multi-Agent LLM Orchestration GenAI Platform
Enterprise GenAI Orchestration Platform

A production-grade multi-agent GenAI platform enabling planning, reasoning, retrieval, validation, and response generation for complex engineering and telecom workflows. Built on LangGraph with stateful execution, memory modules, and a multi-LLM routing layer for dynamic model selection by cost, latency, and task complexity.

Multi-agent orchestration flow
Orchestrator LangGraph stateful graph Planner Agent Task decomposition Retrieval Agent RAG + knowledge Validation Agent Schema + guardrails Response Agent Output generation Multi-LLM Router — AWS Bedrock / OpenAI / Claude
Outcome
Modular plug-and-play architecture reduced new agent onboarding effort by ~60%. LLMOps governance framework established prompt versioning, evaluation pipelines, and guardrails for safe production deployment.
LangGraphAWS BedrockOpenAI APIsFastAPIMLflowKubernetes (EKS)Python
Telecom AI Log Analysis Copilot
AI-Powered Telecom Troubleshooting Copilot

An AI-driven telecom engineering copilot using multi-agent architecture for real-time log analysis, anomaly detection, root cause identification, and recommendation generation. Engineers interact in natural language to query network issues and receive contextual, 3GPP-grounded insights.

Copilot pipeline
Telecom Logs CU/DU / EPC KPIs / PM counters Log Parser Agent Structured events Anomaly Detection Agent Correlation analysis 3GPP RAG Context Spec grounding Root Cause + Recommendation Natural language Elasticsearch obs.
Outcome
Significantly reduced manual debugging effort for network engineers. AI-assisted root cause identification surfaced actionable insights in real time, with full observability via Elasticsearch and Grafana.
AWS BedrockLangGraphElasticsearchVector DBFastAPIGrafanaPython
Multi-Agent Code Migration LLM Orchestration
Polyglot Pipeline — Automated Codebase Translation System

A self-hosted multi-agent system that translates an entire repository from one programming language to another, with build verification, automated test validation, and PR-based human review. Designed for production use with durable, resumable orchestration — runs can take hours and survive restarts.

Architecture overview
ORCHESTRATOR durable state machine · resumable ANALYZER dep graph PLANNER topo waves TRANSLATORS parallel · LLM BUILDER / FIXER Docker sandbox · fix loop (max 3 rounds) fix REVIEWER → PR lint · style · open PR · 🔒 human gate GitLab repo · CI · PRs Prometheus Langfuse
Key design insight
Topological wave ordering — translating utils/ before services/ before app/ — is what keeps cross-file references consistent. It's the single biggest failure mode of naive file-by-file translation, solved by translating tests first, then the code they cover.
TemporalRedisvLLMClaudeDockerGitLabSonarQubeLangfusePrometheusPython
Skills & expertise
Built across 9+ years in production telecom and AI systems.
Generative AI & LLMs
RAG PipelinesAgentic AILangGraphPrompt EngineeringClaudeGPTMistralQwen
Cloud & Infrastructure
AWS BedrockSageMakerEKSS3DockerKubernetesFastAPI
Telecom Systems
5G NRCU/DU Split-63GPP TS/TRvRANEPCMME/SGW/PGW
MLOps / LLMOps
MLflowKubeflowCI/CDEvaluation PipelinesGuardrailsJenkins
Data & Knowledge
Amazon NeptuneVector SearchHybrid RetrievalElasticsearchGraphQL
Languages
PythonC/C++BashLinuxboto3Grafana
Patents
Filed during B.E. at PCCOE — early signals of an inventive engineering mindset.
P1
Smart Boxes Notifying Unavailability of Grocery
Application No. 4488/MUM/2015
P2
Wallet Protection and Monitoring Using Smartphone
Application No. 4614/MUM/2015
P3
Smart Remote Accessing Device (for TV)
Application No. 4615/MUM/2015

Let's build something intelligent

Open to senior AI Architect roles, consulting engagements, and speaking opportunities.