AI Engineer shipping production systems and leading AI transformation, on 18 years of telecom network performance at Huawei

Multi-agent systems on Google Cloud. Solutions architecture across AWS, Azure, and GCP. Scale-to-zero GPU inference on Kubernetes. Fine-tuned edge models and custom MCP servers.

Multi-agent system on Google Cloud that turns a customer complaint into a triaged NOC incident ticket in under 30 seconds

A Google ADK SequentialAgent orchestrates four LlmAgent sub-agents on Gemini (Vertex AI), collapsing the manual NOC workflow of correlating network events, call detail records, and ticketing into one natural-language step. Every LLM call routes through a 4-attempt model-failover ladder shown live in the UI, so a rate-limited model is swapped for the next one on screen instead of failing the request.

• Top 100 of thousands at the Google Cloud Gen AI Academy APAC 2026 (Cohort 1), live on Cloud Run at netpulse-ui.run.app
• Substrate-swappable by design: the BigQuery + AlloyDB pilot now runs zero-idle-cost on bundled SQLite, swappable back through one MCP Toolbox tools.yaml line

Solutions architecture across AWS, Azure, and GCP: the same production workload built 1:1 on all three from a single cloud-neutral contract

One stateless HTTPS FastAPI service over managed PostgreSQL, with autoscaling, private networking, workload identity, and observability, specified once against a cloud-neutral contract and then implemented identically on AWS, Azure, and GCP. The deliverable is the comparison itself: a defensible three-cloud Bill of Quantities with every line-item cost delta explained, plus a per-cloud evidence pack that proves each requirement on live infrastructure. Built end to end as a self-directed, hands-on study.

• Defensible cost comparison: a like-for-like three-cloud Bill of Quantities in one region, with every cost delta traced to a specific service line
• Keyless by construction: workload identity on each cloud, checked against a V1-V9 acceptance rubric run on live infra and captured as a screenshot evidence pack

Scale-to-zero GPU inference on Kubernetes that costs $0 when idle

Two-layer autoscaling: KEDA watches Redis queue depth for 0-to-N pod scaling while the GKE Cluster Autoscaler provisions and deprovisions GPU VMs on pending pod scheduling. vLLM serves with continuous batching; model weights persist on a PVC and image layers pre-cache via Secondary Boot Disk, cutting cold start 48% (11 to 5.6 min).

• True scale-to-zero at both the pod and node level, $0/hr when idle
• Full observability: 12-panel Grafana dashboard (Prometheus + NVIDIA DCGM) plus Locust load testing

A fine-tuned 270M edge model that beats generalist function-callers by baking tool knowledge into its weights

LoRA (r=128) fine-tuning on Gemma 3 270M moves all 14 tool definitions out of the prompt and into the weights: a ~20-token query goes in, a JSON tool call comes out, regardless of tool count. The result is 99.5% accuracy at 32x fewer prompt tokens than schema-in-prompt baselines, in a 291 MB Q8_0 GGUF that runs on phones, laptops, and Raspberry Pi.

• Trained on a single consumer GPU (RTX 4060, 8GB VRAM): 418/420 eval accuracy from 14,033 training examples
• Fully local, zero API cost: 14 tools across 2 MCP servers at 153ms average latency

A custom MCP server behind a Telegram bot: snap a receipt, get live project cost accountability

Field workers photograph receipts in Telegram; an LLM-vision agent runs OCR to pull amount, vendor, and date, confirms over inline buttons, and writes a per-submitter expense to a SQLite ledger exposed through a custom MCP server. Managers query live budget-vs-actual and get matplotlib charts back in-chat. Built for a real telecom subcontractor whose paper-and-form expense tracking kept failing.

• Custom 6-tool MCP server (set_budget, log_expense, query_spend, budget_status, budget_chart, export_ledger) over an async SQLite ledger, with a 37-test pytest suite
• Self-hosted agent: OpenClaw runtime on Azure OpenAI vision, Docker Compose deployment; the Azure pilot is archived and production has since moved to AWS

Six end-to-end telecom ML projects, each on synthetic data with embedded network physics

Six self-contained projects spanning classification, regression, time-series, and reinforcement learning. Most ML demos treat telecom as generic tabular data; here each project hand-crafts synthetic data with embedded telecom physics (temporal correlation, spatial clustering, equipment failure signatures), then runs the full paradigm end-to-end from data generation through feature engineering, training, evaluation, and business-insight translation. The repo is an index; source lives in six child repos.

• Supervised: Churn AUROC 0.86, RCA Acc@1 0.91, QoE RMSE 0.45, Capacity MAPE 14.5%
• Unsupervised + RL: Anomaly F1 0.70, Network Optimization +61% vs random

• Location: Jakarta, Indonesia (UTC+7), open to remote
• LinkedIn: linkedin.com/in/adityonugrohoid
• Email: adityo.nugroho.id@gmail.com