# GOKULNATH V ## AI Engineer **Bengaluru, India** | [gokulnathvwork@gmail.com](mailto:gokulnathvwork@gmail.com) | +91 9108107085 | [gokulnathv.is-a.dev](https://gokulnathv.is-a.dev) | [LinkedIn](https://www.linkedin.com/in/gokulnath-v-2003g/) | [GitHub](https://github.com/Coding-Devil) --- ## SUMMARY AI Engineer at Nokia architecting and shipping production-grade **agentic AI systems** for autonomous 5G Core operations. I specialize in transforming AI intelligence into reliable software through **agent harnesses, context-engineering, MCP/A2A orchestration, AgentOps, advanced RAG and graph-grounded reasoning**. Strong hands-on in the **Python AI ecosystem**, with a robust track record of delivering production-ready multi-agent systems, hybrid RAG pipelines and full-stack AI compliance tools. Graduated as the Best Outgoing Student and recipient of multiple Best Project Awards. --- ## TECHNICAL SKILLS - **Agentic AI Systems:** Multi-agent orchestration, agent harnesses, MCP, A2A, context engineering. - **LLM Application Engineering:** LangGraph, LangChain, CrewAI, Pydantic AI, OpenAI SDK, model routing & serving, Hugging Face Transformers, PyTorch, Ollama, vLLM, Unsloth. - **Retrieval & RAG Systems:** Vector-RAG, GraphRAG, PageIndex, BM25, Reciprocal Rank Fusion, Semantic caching. - **Evaluation, AgentOps & Reliability:** Opik, LangFuse, TTFT / latency analytics - **Knowledge Graphs & Semantic Data:** RDF, OWL2, SHACL, SPARQL, GraphDB (Ontotext), Neo4j, Cypher, ontology engineering. - **Data Stores & Vector Infrastructure:** ChromaDB, Qdrant, Redis / RedisVL, PostgreSQL, Supabase, pgvector, indexes, metadata filtering, embedding pipelines. - **Cloud-Native & Observability:** Docker, Kubernetes, OpenShift, AWS foundations, Vercel, GitHub Actions, Prometheus, Grafana. - **AI Programming:** Python, FastAPI SQL, React, Next.js - **Model Optimization:** LoRA, QLoRA, RLHF / DPO concepts, Post-Training quantization (GGUF & bitsandbytes) --- ## EXPERIENCE ### Nokia — AI Engineer **Technology Architecture & Partnerships Office | Bengaluru, India | July 2025 – Present** - **Leading the development of production-grade agentic AI systems** for autonomous operations and runtime intelligence inside **5G Core Networks** – owning end-to-end design across **multi-agent workflows, RAG pipelines, MCP & A2A orchestration, and custom reasoning engines** for high-stakes, trustworthy decision-making. - **Architecting deterministic agent harness mechanisms** that reinforce LLM-driven decisions with robust **context engineering**, dynamic execution contracts, and quality-gated guardrails, enabling precise operational state awareness and deeply explainable reasoning under enterprise-grade safety constraints. - **Designed the evaluation and AgentOps layer** powering all shipped agentic systems – LangFuse-style trace lineage, TTFT-latency analytics, deterministic input-output logging, and regression checks, so agent behavior is measurable, debuggable, and audit-safe. - **Engineering Semantic reasoning systems using RDF/OWL2 ontologies, SHACL constraints, and graph-grounded inference engines** to give agents a single source-of-truth of the operational domain, accelerating zero-touch operations across cloud-native telco workloads. - **Built the Hybrid KG-RAG retrieval engine** combining **Multi-Query embedding, HyDE, BM25, RRF and graph-based context expansion** – now adopted as the shared retrieval backbone for multiple internal copilots, delivering cited, grounded answers over Target-architecture documents for compliance and policy insights. - **Partnering with product portfolio teams** to embed trustworthy AI components into Cloud-Native Network Functions (CNFs) on Kubernetes / OpenShift and driving internal R&D analysis across emerging AI stacks, co-authoring **reference architectures, whitepapers and AI capability blueprints** to be reused across Core product groups to standardize AI adoption. - **Owned and spearheaded the zero-to-one delivery of the flagship Self-X Agentic Core** – a multi-agent system autonomously powering intent-driven monitoring, protection, and version-upgrade for 5G Core networks. Feature impact recognized at **Mobile World Congress 2026 (Spain)** and the **Core-User Group Summit (Munich, 2026)**, demonstrating next-generation Autonomous-Networks. ### Nokia — R&D Intern (AI) **Cloud & Network Solutions — Product & Engineering | Bengaluru, India | Jan 2025 - Jun 2025** - **Contributed to early R&D on Semantic & Agentic Stack** by evaluating multi-agent frameworks, Model selection and graph representations (LPG vs. RDF), then modelling **3GPP 5G network domains as semantic knowledge graphs** using **RDF/OWL2 and GraphDB** to support explainable planning and operational reasoning. - **Built the Agentic Network Planner** — a chat-driven, KG-grounded planning assistant that turns natural-language intent into validated 5G topologies with safe two-way semantic database interaction. ### Nokia NBUC — Undergraduate Project Lead **Bengaluru, India | May 2024 - Nov 2024** - Led a 4-engineer team to design and implement a **Reinforcement Learning** framework for 6G base-station self-optimization, focused on real-time QoS and resource allocation. - Achieved a **30% latency reduction** in dynamic parameter-tuning experiments while preserving resource usage and QoS targets, validated across simulator scenarios. - Won **Second Best Student Project** among top universities at **Nokia NBUC 2024**; work contributed to 6G patent-oriented ideation. ### Samsung Innovation Campus — AI Trainee **Bengaluru, India | Jan 2024 - Jun 2024** - Completed Samsung’s intensive AI training across **ML, deep learning, computer vision, NLP, and tabular modeling**, graduating with high merit. - Built mentor-reviewed AI projects and qualified as a **regional finalist** in the Samsung program. --- ## SELECTED PROJECTS ### Agentic Core — Self-X Multi-Agent System for 5G Core **LangGraph, MCP, Prometheus, Grafana, Kubernetes/OpenShift** - Engineered the flagship **Self-X agent suite** for autonomous 5G Core operations—transforming manual monitoring, protection, and upgrade workflows into fully operator-armed, autonomous flows across cloud-native network functions—by designing four specialized, bounded agent roles per CNF :**Self-Monitoring, Self-Healing** (both reactive and proactive), **Self-Config**, and Implemented symmetric **A2A peer coordination** with **Self-Orchestration** agent — acting as the only external communication boundary. - Architected **context-driven orchestration** in which dynamic context management forms the operational backbone—ensuring agents ground their decisions in holistic knowledge of evolving system state, role boundaries, and mission objectives (beyond just raw telemetry). This operational context layer enforces reliability, trust, and transparent, intention-aware interactions between agents and their environment. - Designed a transparent, end-to-end traceability architecture by publishing agent-milestone events to a shared knowledge graph—meticulously logging every agentic decision point, rationale, and action trigger. Presented these summaries in **Grafana-embedded dashboards** for operator audit and trace review. ### Hybrid KG-RAG — ULTRA Retrieval Engine **Langchain, ChromaDB, Neo4j, BM25, Cypher, FastAPI, Streamlit** - Built a reusable **enterprise RAG engine** to answer architecture and policy questions from large technical documents with citations, retrieval traces, and source-grounded responses. - Combined **Multi-Query expansion, HyDE, BM25, dense vector search, Reciprocal Rank Fusion and optional LLM reranking** to improve recall beyond single-vector retrieval. - Implemented **parent-child chunking** (1024-token parents, 128-token children), allowing precise child-chunk search (in vectorDB) while returning richer parent context (from Knowledge Graph) for answer generation. - Added **page graph expansion** through a page-centric Neo4j graph, preserving neighboring-page context and reducing brittle answers at document boundaries. - Exposed **Turbo (~10s)** and **Full (~25s)** modes with relevance scores, source citations, and full retrieval traces; reused as the retrieval backbone for downstream copilots over enterprise architecture documents. ### CNF Compliance Copilot — Hybrid SHACL + KG-RAG **OWL2 ontologies, pySHACL, D3.js, RDFLib, Qdrant** - Built a full-stack compliance copilot that converts Kubernetes / CNF manifest review into a deterministic, explainable validation workflow. Integrated the **RAG engine** to explain each violation with page-cited answers, turning compliance failures into actionable remediation guidance. - Engineered a **hybrid YAML → RDF → SHACL** pipeline that maps raw manifests into traceable RDF graphs, then validates them against target-architecture policies. - Delivered a **React + D3 force-directed RDF graph viewer** plus Markdown / JSON reports with severity classification, RDF traceability, and fix suggestions. ### Agentic Network Planner — Chat-to-Topology Planner **React, Stack-AI, Ontotext GraphDB, RDF / SPARQL, Cytoscape, OpenStreetMap** - Built a chat-driven planning assistant that converts natural-language network design intent into **validated, KG-grounded 5G topology proposals** across regions, sites, clusters, and network functions. - Engineered the planner to leverage **SparQL templates as tools** for topology generation that faithfully reflects existing assets, capacities, and ontology constraints, empowering agents with safe, two-way semantic database interaction—ensuring all designs are grounded in real, validated knowledge rather than free-form output or hallucinated ontology terms. - Rendered generated topologies for human-in-the-loop visual review, preserving operator control over final network design choices. ### Universal Installer AI Assistant — SaaS Deployment Copilot **SaaS Platform Integrations, AWS Bedrock, OpenAI Agent-SDK** - Built an AI-assisted deployment copilot embedded inside a SaaS installer MVP for complex on-prem cloud-native rollouts. - Combined **guided Q&A flows, golden configuration rules, and JSON schema validation** to prefill audit-friendly deployment blueprints and reduce manual configuration errors. - Connected user design intent to downstream deployment automation across infra-node, hub-cluster, and workload-cluster stages and helped to make installation workflows more self-service by explaining configuration choices, defaults, and downstream impact inside the installer experience. ### Side Projects (Selected) - **MoneyMates** — Family finance OS with transaction insights, budget-tracking, billing-reminders using Supabase/PostgreSQL persistence, and a polished React + TypeScript dashboard, with Chatbot inetgration featuring Text-to-SQL capability. - **OpenFootball RAG** — Built a Football RAG system for FIFA World Cup data with semantic embedding retrieval (ChromaDB) and fast scoring (FlashRank). Tuned for precise results & scalable datasets. - **Second Brain Telegram Bot** — Zero-cost private knowledge assistant over personal notes, using Telegram bot, Supabase + pgvector, local CPU embeddings (`all-MiniLM-L6-v2`) hosted on Render and model routing via OpenRouter - **BRUTUS Utility** — Engineered a local-first, offline-capable productivity PWA featuring Pomodoro timer, task management, note-taking, and garage inventory modules. Architected all data persistence in `localStorage` to ensure strict client-side data residency by design. --- ## EDUCATION ### RNS Institute of Technology, Bengaluru **B.E. in Computer Science (Data Science) | 2021 - 2025** - **CGPA: 9.45 / 10** — **Institute Rank 1**, **University Rank 4** - Awards: **Best Outgoing Student of the Institution**, **Medal of Merit** --- ## CERTIFICATIONS - **Machine Learning Specialization** — Stanford University / DeepLearning.AI - **IBM AI Engineering Professional Certificate** — IBM - **IBM Agentic AI & RAG: AI Workflow in Production** — IBM - **Artificial Intelligence Course** — Samsung Innovation Campus - **AWS Cloud Foundations** — Amazon AWS Academy - **Google Data Analytics Professional Certificate** — Google