Projects

Consumer Android app for Filipinos. On-device AI receipt OCR (ML Kit + Qwen 2.5 / SmolLM), SALN module for government employees with CSC-template DOCX/PDF export, AI Planner with optional Bring-Your-Own AI endpoint, encrypted vault export (AES-256-GCM), and Aninaya Bridge for phone-to-phone LAN sync. AES-256 SQLCipher database. Published on Google Play.

Enterprise RAG-based document extraction engine with parallel batch processing (14-step pipeline), multi-provider AI support (Ollama, OpenAI, Anthropic, Gemini, Groq), Human-in-the-Loop review workflows with active learning, and multi-tenant RBAC. In a production proof-of-concept it reached 99.95% cell-level and 99.40% row-level extraction accuracy across ~22,000 fields on structured government enrollment forms. Deployed internally at DSWD Philippines.

Voice + text multi-agent orchestrator with a 3-ring, offline-first architecture. Local wake-word ("hey atlas") via openWakeWord, faster-whisper STT and Piper TTS, with on-device Ollama models for intent routing and a Claude/Gemini cloud tier for hard reasoning. Retrieval over a personal vault and diary using sqlite-vec, plus a JARVIS-style HUD built with FastAPI + HTMX.

Real-time IoT monitoring platform with MQTT ingestion, ESP32/Wemos firmware, TimescaleDB time-series storage, automated alerting, and multi-tenant farm management. Originated at UPLB (2018-2019), revived and modernized.

Document-automation pipeline for semi-monthly government reporting. Converts Markdown + YAML frontmatter into submission-ready Word, PDF, and Google Docs output via direct python-docx / OOXML editing — reproducing official templates faithfully (column widths, merged headers, signature blocks, live page fields). Multi-format export with a guided CLI wizard.

Local-first Flutter BIR tax estimator for Philippine freelancers and mixed-income earners. 8% vs graduated tax comparison, 1701Q quarterly worksheet, and a BIR deadline tracker — sharing an on-device data core with the Aninaya finance app so figures never need re-entry. In development.

Privacy-first n8n automation companion for Aninaya. Uses a pull model over a secure HTTPS GET endpoint (not inbound push) to ingest transactions from email, keeping credentials and data under the user’s control. MIT-licensed.

Local LLM-powered assistant for business search. Uses Ollama for on-prem inference with FAISS vector similarity search. Demonstrates data sovereignty and zero-cost AI inference.

Enterprise Resource Planning system built on the Frappe framework with AI-powered task summarization using Ollama. Integrates LLM capabilities into business workflow management.

Collection of self-hosted n8n workflow automations running via Docker. Demonstrates workflow design, API integration, and process automation patterns.

Open-source Human Resource Management System with payroll processing, employee records, and attendance tracking built on the Frappe framework.