Reynald Ace Pilpil

A 9-step agent with integrity enforcement — architecturally prohibited from fabricating resume content. A post-tailoring diff check flags violations, with every change logged in a fully auditable changelog (reason + confidence score). Pipeline: resume parsing → JD scoring → company research → ATS-tailored resume + cover letter + outreach in under 30 seconds.

An autonomous research agent with a self-evaluating revision loop using LangGraph — agent critiques its own draft, scores it, and re-searches targeted gaps until quality threshold is met. Researches a company across 4 parallel angles (overview, news, competitors, market trends) via live web search and delivers a structured intelligence report as a Word document.

End-to-end RAG pipeline over the Philippine Labor Code — document ingestion, semantic chunking, vector embeddings (all-MiniLM-L6-v2), ChromaDB retrieval, and Flan-T5 generation. Returns natural-language answers (leave entitlements, overtime rules, termination procedures) with source citations grounded in the actual law, reducing hallucination through retrieval-first design.

Production-grade REST API that ingests bank statement PDFs and returns structured JSON — extracted transactions, auto-categorized spending, computed financial summaries (income, expenses, savings rate), anomaly flags, and actionable insights. Containerized with Docker, deployed to Railway with auto-generated Swagger / OpenAPI docs.

Extracts structured data (menu items, quantities, prices, totals) from receipt images using LayoutLMv3, a multimodal transformer that jointly reasons over text, layout, and visual features. Achieved 93.96% F1 on validation set, demonstrating document AI capability beyond plain-text NLP.

Multi-input bulk sentiment analysis tool that processes batches of reviews, tweets, or comments and returns an aggregate sentiment score, breakdown chart, and the top positive / negative phrases driving the score. Dashboard-style output (not single-label classification), making it usable as a social-media intelligence tool.

Production-ready churn risk dashboard: ingests CSV customer data, predicts churn probability with XGBoost, and outputs a retention report with risk distribution, ranked high-risk customers, and SHAP-based feature explanations — translating raw model output into business-actionable retention recommendations using explainable AI (SHAP).