Building production AI systems β LLM agents, statistical analytics platforms,
and full-stack applications with real machine learning at the core.
I design and ship end-to-end AI-powered systems β from the machine learning layer through the backend API to the frontend interface. My work sits at the intersection of applied ML, LLM engineering, and decentralized infrastructure.
Most recently I built Subnet Sentinel, a full-stack Bittensor intelligence platform covering 128 active subnets. It replaced hardcoded threshold alerts with a per-subnet statistical anomaly detection engine (z-score baselines, compound multi-metric detection, 4-tier severity taxonomy), added a RAG-based conversational AI agent with multi-turn session memory and tool-first groundedness, and a local LLM pipeline (Ollama + Qwen3 8B) for deep GitHub repository analysis β all without a GPU.
π Subnet Sentinel β Bittensor Intelligence Platform
Full-stack AI platform monitoring 128 Bittensor subnets in real time
- Statistical anomaly detection β per-subnet rolling baselines (mean, std, z-score, p10/p90) over 30-day snapshot history. Alerts only when a metric deviates from that subnet's own historical norm β not a global threshold
- Compound anomaly detection β fires a single high-confidence alert when β₯2 metrics deviate simultaneously (e.g. emission drops 3.1Ο AND miner count drops 2.8Ο)
- RAG conversational agent β Gemini + Chroma vector store, multi-turn session memory, SSE streaming, tool-first groundedness (no number stated without a live tool call backing it), freshness disclosure per answer
- Local LLM pipeline β map-reduce GitHub repository analysis using Ollama (Qwen3 8B, CPU-only): shallow-clone β filter high-signal files β per-file summarization β structured reduce into miner/validator requirement profiles
- Hybrid LLM routing β Ollama (local) for background summarization, Gemini for interactive chat β switchable per component via config
- On-chain change detection β tracks registration state, emission transitions, burn rate flips, miner/validator churn across all 128 subnets with deduplication
- 4-tier severity taxonomy β LOW / MEDIUM / HIGH / NEWSLETTER with operational impact descriptions and
requires_actionflags
Python FastAPI Next.js PostgreSQL Chroma Ollama Gemini API Bittensor SDK Docker APScheduler
π‘οΈ Tensorprox β Decentralized DDoS Detection (Bittensor SN91)
ML-powered cybersecurity node deployed on the Bittensor network
- Sourced and engineered a DDoS traffic dataset from scratch (no existing dataset for this use case)
- Trained a Random Forest classifier reaching 98% classification accuracy distinguishing malicious from normal traffic
- Built the full pipeline: synthetic attack-traffic generation β feature engineering β model training β live inference integration
- Deployed as a production miner node on Bittensor Subnet 91
Python scikit-learn Random Forest AWS GCP Bittensor
Applied ML systems across reinforcement learning, forecasting, and optimization
| Subnet | Type | Tech |
|---|---|---|
| Precog (SN55) | Bitcoin price-prediction model | Time-series forecasting, ML |
| Swarm (SN124) | Autonomous drone flight agents | Stable Baselines3, PyTorch, JAX |
| Sturdy (SN10) | Yield optimization across DeFi protocols | Uniswap v3, Aave, Compound v3, Morpho |
| Liquidity Auction (SN77) | Voting system with Ed25519 signature verification | RESTful API, Token Economics |
| SubVortex (SN7) | Subtensor node infrastructure | Node Operations |
| Investing (SN88) | TAO/Alpha staking strategy design | Staking, Token Economics |
Python PyTorch Stable Baselines3 JAX scikit-learn Blockchain
Solidity-based on-chain system for a GameFi ecosystem (Singapore)
- Architected 4 core smart contract modules (Factory, Blueprint, Product, Custody) across 3 token standards (ERC-20/721/1155)
- Built FactoryGame: NFT-synthesis platform with factory contract deployment system for modular, scalable asset creation
- Built Stakeverses: staking platform with reward formula design, staking/claiming/withdrawal contracts supporting 3 reward token types
- Security-audited all contracts with Hardhat, Chai/Mocha, Sepolia testnet, and the Aderyn audit tool before deployment
- Web3 frontend: Ethers.js, Wagmi, RainbowKit, WalletConnect v2 for multi-wallet support
Solidity ERC-20/721/1155 Hardhat Aderyn TypeScript React Node.js GraphQL MongoDB
NFT marketplace with auction and royalty mechanics
- Built NFT collection and marketing contracts including auction and royalty mechanisms
- Concept announced at the Agora "Celebrating NFT & DeFi" event (June 2021)
- Publicly launched on stage at GITEX Dubai (Oct 2021) in front of a 1M+ audience
Solidity Web3.js React ERC-721
Techniques: Random Forest Β· Reinforcement Learning (Stable Baselines3, JAX) Β· Time-Series Forecasting Β· Statistical Anomaly Detection (z-score, rolling baselines) Β· RAG Β· LLM Agent Orchestration Β· Prompt Engineering Β· NLP Β· Transformer-based Models Β· Local LLM (Ollama, Qwen3)
Protocols: ERC-20 / ERC-721 / ERC-1155 / ERC-4626 Β· DeFi Β· NFT Β· DAO Β· Bittensor Β· Wagmi Β· RainbowKit Β· WalletConnect v2
2026 β Present Subnet Sentinel (Personal Project) β AI platform engineer
2026 β Present LiveSpan Network Inc. β Founding Engineer (edge computing platform)
2025 β 2026 Upwork β Freelance AI Full-Stack Developer
Β· BNB Tweet (OpenAI GPT pipeline, 3-format export)
Β· WealthLogix (Django, PDF generation, 100% accuracy)
2025 Bittensor β AI/ML Engineer (7 concurrent subnets)
2023 β 2025 FactoryVerse β Architecture Designer, Smart Contract Developer
2021 β 2023 TwistedResources β Web Developer
2020 β 2021 TwistedResources β Intern Web Developer
π B.S. Computer Science β Holy Angel University, Philippines (2017β2021)
π Champion, Dapp and Dash Campaign β Forward, Rootstock on AlphaGuild (NovβDec 2023)
π€ GITEX Dubai Stage Launch β Presented Sinverse Marketplace to 1M+ audience (Oct 2021)
Open to full-time and contract AI/full-stack engineering roles β remote