Adityo Nugroho

AI/ML Engineer - Telecom/Industrial Domain Expert | 18+ Years Network Optimization Leadership (Deputy GM at Huawei)

I build AI systems to address real operational bottlenecks, from event-driven log analysis with 63x noise reduction to RAG-based troubleshooting and agentic automation with explicit safety gates.

GitHub LinkedIn Email

01. About Me

After leading network optimization strategy for operators serving millions, I witnessed what breaks at scale and built AI systems to fix it. I pivoted into AI/ML engineering to address these bottlenecks, architecting and deploying 12 end-to-end systems in 14 months (Nov 2024-Jan 2026), including AI/ML systems (GenAI, RAG, Agentic AI, MLOps, AI Observability) and production infrastructure.

"I translate operational complexity into measurable outcomes by designing AI systems that remain reliable under real-world stress, not just controlled demonstrations."

02. Featured Projects

Gemini 2.0 Flash Lite • Async Python • RAG • ChromaDB • Function Calling • Pydantic

TRINITY Operations Suite

A cohesive integrated suite solving the "Mean Time to Recovery" bottleneck with three integrated components deployed as part of NOC operations:

  • Incident Commander: Event-driven log analyzer with tumbling window batching (5s or 100 items) and async architecture for non-blocking I/O. Uses semantic root cause clustering via Gemini 2.0 Flash Lite with Pydantic-enforced structured outputs for real-time incident detection.
  • NOC-Oracle: RAG-powered troubleshooting with hybrid search combining semantic vector search and keyword boosting for exact code matching. Features context-aware chunking preserving error-solution relationships and strict context enforcement preventing hallucination.
  • Net-Ops Agent: Agentic AI with reasoning-action separation pattern and human-in-the-loop approval gates. Uses deterministic function calling from pre-defined toolbelt with Pydantic validation ensuring type safety. All actions require explicit authorization.
Vertex AI • Streamlit • Cloud Run • Gemini 2.0/2.5 • Docker

Google Cloud AI Studio ● LIVE

Cloud-native Streamlit application demonstrating advanced GenAI workflow orchestration using Google Vertex AI. Transforms simple text descriptions into photorealistic architectural renders through a 3-stage pipeline: text enhancement → sketch generation → photorealistic rendering.

  • Text Enhancement Stage: Uses Gemini 2.0 Flash Lite to expand simple user descriptions into detailed architectural specifications, preparing enriched context for downstream image generation stages
  • Sketch Generation Stage: Gemini 2.5 Flash Image creates architectural line drawings from enhanced text descriptions, creating intermediate visual representations for rendering pipeline
  • Photorealistic Rendering Stage: Transforms sketches into V-Ray style photorealistic renders using advanced prompting techniques for realistic architectural visualization
  • Cloud Run Deployment: Containerized Streamlit application deployed on Google Cloud Run with Vertex AI integration and Docker containerization for scalable cloud deployment
Parquet • XGBoost • LightGBM • SHAP • Optuna • Deterministic Seeding

Telecom Digital Twin + MLOps Pipeline

A complete MLOps ecosystem combining deterministic data generation with ML workflows. Generates 50K users with 5.6M sessions in reproducible fashion, then applies six-phase analytics pipeline from EDA to strategic recommendations.

  • Digital Twin: Deterministic multi-table generator with bit-exact reproducibility via cascade-based seeding. Produces 4 output tables (users, cells, sessions, events) with referential integrity validation. Parquet columnar storage with embedded schema metadata for efficient storage.
  • QoE Analytics: Six-phase analytics pipeline achieving R²=0.7247 (XGBoost, MAE=0.3672, RMSE=0.4560) and ROC-AUC=0.9645 (LightGBM, precision=0.46, recall=0.92). Features SHAP interpretability for model explainability and Cohen's d effect size analysis for quantifying impact magnitude of QoE drivers.
Async Python • WebSockets • Donchian Channels • Systemd • Ed25519

Trailing Edge Trading Bot

High-performance asynchronous Python trading bot for Binance featuring dynamic trailing take-profit strategies, regime detection, Donchian channel gating, and Ed25519 authentication. Built for 24/7 autonomous operation on low-cost infrastructure with low-latency reaction times.

  • Dynamic Trailing Take-Profit: Exponential decay from start to min factor, automatically adjusting profit targets as gains increase, locking in profits while allowing upside
  • Regime Detection and Auto-Compounding: Automatic switching between BASE (holding inventory) and QUOTE (holding cash) modes with all-in compounding for bull rallies
  • Donchian Channel Hard Stop: Volatility-aware stop-loss using Donchian Channels with intelligent re-entry gating that waits for price to cross mid-channel in favorable direction after stop triggers
  • 24/7 Systemd Deployment: Deployed as systemd service for autonomous operation with auto-restart, resource limits, and journald logging
Multi-Chain • FIX 4.4 • Moralis • Ankr • Binance

RATU Trading Suite

A comprehensive 4-component crypto trading infrastructure for multi-chain token discovery, on-chain analytics, market data, and low-latency order execution:

  • FIX Bot: FIX Protocol 4.4 connector with ED25519 authentication and defensive message parsing. Features three FIX sessions (Market Data, Order Entry, Drop Copy) with spread market making strategy for low-latency order execution
  • REST API: Comprehensive market analytics using 7 Binance public endpoints with multi-timeframe kline analysis (1h, 4h, 1d) and connection pooling. Dataclass-based response parsing ensures type safety
  • On-Chain Monitor: Token holder analytics and whale tracking across 6 chains (BSC, Ethereum, Polygon, Arbitrum, Base, Avalanche) with three-mode architecture (Basic Info, Top Holders, Full Snapshot) and known label identification
  • Moon Radar: Multi-chain DEX pair scanner supporting 4 chains (Ethereum, BNB Chain, Polygon, Base) with dynamic pair parsing and early token detection via 24h price gains
XGBoost • LightGBM • SHAP • Optuna • Python 3.11

Telecom ML Framework

A framework for building AI/ML solutions to real-world telecom challenges, emphasizing domain expertise and practical problem-solving. Provides 6 ML project templates covering the most common telecom AI/ML use cases with complete technical specifications.

  • Six Use Case Specifications: Complete specs for Churn Prediction, Root Cause Analysis, Anomaly Detection, QoE Prediction, Capacity Forecasting, Network Optimization with problem framing and model recommendations
  • Domain-Informed Data Generators: Hand-crafted generators embedding real telecom physics (SINR, Shannon capacity, congestion patterns) vs generic synthetic data, demonstrating domain expertise
  • SHAP-Compatible Standards: Enforced dependency versions (numpy<2.0, xgboost<2.0) ensuring interpretability works without conflicts, critical for business adoption
  • Unified Project Structure: Consistent template across all projects with config, data generation, features, models, notebooks, and tests enabling rapid project creation
View Complete Portfolio

03. Technical Stack

Programming & Frameworks

Python, Python Async (asyncio), Streamlit, FastAPI, LangChain, Pandas

AI & Machine Learning

Google Gemini 2.0 (Flash/Lite/Pro), Vertex AI, RAG, XGBoost, LightGBM, scikit-learn, SHAP, Agentic Patterns, Function Calling, Prompt Engineering

MLOps & Data Engineering

Reproducible Pipelines, Schema Validation, Model Evaluation, Parquet, Synthetic Data (SDV), Text Embeddings, Batch Prediction, CLI Automation, SQL

Cloud & Infrastructure

Google Cloud Run, Docker, Artifact Registry, Linux VPS, Systemd, CI/CD

Protocols & APIs

REST, WebSockets, FIX 4.4, JSON-RPC, GraphQL

Tools

UV Manager, Pytest, Ruff, Git/GitHub, Tableau

04. What's Next?

I am currently open for opportunities!

Available for AI/ML Engineering roles focusing on:

  • Telecom/Industrial AI: Domain-specific AI/ML solutions leveraging network optimization expertise
  • Observability & APM: Automated remediation and log analysis
  • Fintech & Trading: Low-latency decision systems
Get in Touch