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 AIRAG PipelinesLLM Orchestration5G NR / 3GPPAWS BedrockLangGraphHealthcare AI3 Patents
All architectures are conceptual representations of real production systems. No proprietary code or data is shared.
RAGKnowledge SystemsTelecom
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
Outcome
Reduced 3GPP specification query resolution from ~45 minutes of manual search to under 30 seconds, deployed across the engineering team.
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
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.
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
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.
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
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.