I design and build AI agents and autonomous pipelines that replace hours of manual work with reliable, self-running software, multi-provider LLM routing, RAG, tool-calling, and local/on-device inference. I don't ship demos, I ship systems that run.

My edge: a real engineering foundation, a Computer Engineering degree and systems work from Python to embedded C, so the agents I build are debugged, secured, and maintainable, not prompt-and-pray.

🎓 Computer Engineering, La Salle University – Ozamiz (2026)  |  📍 Philippines (UTC+8), available for remote work worldwide 🌐 Portfolio  |  📧 jasper.paitan0918@gmail.com  |  LinkedIn  |  Credly Certifications

• AI Agents & Autonomous Workflows — agents that scrape, reason, decide, and act end-to-end, with tiered human-in-the-loop control
• Multi-Provider LLM Systems — routing, failover, and key rotation across Groq, Claude, OpenAI, OpenRouter, and Cerebras for reliability and cost control
• RAG & Local AI — private, on-device assistants (Ollama, ChromaDB) that keep data on the client's own machine
• Business Process Automation — turning repetitive, multi-step operations into scheduled, hands-free pipelines
• Applied ML & Edge AI — computer vision and real-time inference on-device (TensorFlow, Edge Impulse, Raspberry Pi)

I help teams find the highest-ROI workflow to automate, then design, build, and hand over the agent that does it, documented, reliable, and yours to keep. Available for remote consulting engagements and agentic / AI engineering roles.

• Automate a manual, repetitive workflow (research, outreach, data entry, reporting) with an autonomous agent
• Add an AI layer (chat, RAG, classification, routing) to an existing product or ops stack
• Stand up a private, local LLM setup so sensitive data never leaves your infrastructure
• Build your small business a premium website wired for lead response — form to instant reply to follow-up, so no inquiry goes cold

A single-file, zero-backend AI coding assistant for game developers, engineered for reliability across multiple LLM providers.

• Multi-provider routing with automatic failover and API-key rotation across Groq, OpenRouter, and Cerebras, so a rate-limited or down provider never stops the session
• Multi-round tool-calling: the agent calls tools, reads results, and keeps reasoning across steps in one turn
• Two-layer validation pipeline (static analysis + independent AI self-review) that catches and corrects logic errors before code is shown
• Integrated Tavily web search and AI image generation, hardened against XSS, all in one portable HTML file

Tech: HTML/CSS/JavaScript, Groq API, OpenRouter, Cerebras, Tavily Search API

Three vertical demo sites — dental clinic, real estate, home services — each with its own complete design system and a lead-qualifying form. Work samples for my web design + lead-response service.

• Three distinct design systems, hand-built in a single file each: no frameworks, no stock photos, all SVG illustration
• Lead-machine patterns throughout: reply-time promises, qualifying forms (service, budget, urgency), call CTAs
• Live: Dental Clinic · Real Estate · Home Services

Tech: HTML/CSS/JS, design systems, SVG illustration, conversion copywriting (source private)

A private AI agent that runs entirely on-device, no cloud, no data leaving the machine.

• Runs fully local via Ollama — no subscriptions, no cloud APIs, no data leaving your machine
• RAG pipeline (ChromaDB + sentence-transformers) trained on personal notes to answer and draft in your own voice
• Reads and summarizes Gmail, drafts replies in your tone, scans for interview responses
• Autonomously searches and applies to jobs, and scaffolds complete working projects (plus a demo video) from a one-line prompt

Tech: Python, Ollama, ChromaDB, Streamlit, Gmail IMAP/SMTP, GitHub REST API, RAG

An end-to-end agent that scrapes, screens, applies, and monitors replies, with a human kept in control.

• Result: reduced daily job-search time from 2–4 hours to under 10 minutes
• Scrapes listings (Playwright), then scores each with an LLM for fit and scam detection (keyword filter + deep analysis)
• Tiered decision engine (auto-apply / human review / reject) based on AI fit scoring
• Generates a tailored cover letter per role, applies, and classifies inbox replies via IMAP; FastAPI dashboard for live monitoring

Tech: Python, Playwright, FastAPI, SQLite, Claude AI, Gmail SMTP/IMAP

Also rebuilt as a free, zero-cost version using only open-source tools, same pipeline, no paid APIs, to prove cost-efficient automation on any budget.

A real-time computer-vision system that classifies and physically sorts plastics, running entirely on-device with no cloud dependency.

• MobileNet CNN (TensorFlow + Edge Impulse) classifying 5 plastic resin types at greater than 90% accuracy
• Real-time inference fully on-device on a Raspberry Pi 5 — no cloud round-trip
• Servo actuators driven by model output automate the physical sorting end-to-end

Tech: Python, TensorFlow, TensorFlow Lite, Edge Impulse, OpenCV, Raspberry Pi 5

Auto-classifies and files documents using zero-shot NLP and computer vision, no training data, no manual rules.

• Classifies documents by content using zero-shot NLP (facebook/bart-large-mnli)
• Recognizes image types with OpenCV scene recognition; perceptual-hash duplicate detection
• Runs hands-free on a schedule with dry-run and watch modes; SQLite history prevents re-processing

Tech: Python, HuggingFace Transformers, OpenCV, SQLite, schedule

Computer Engineering graduate with a software development internship at Benpos Systems (system maintenance, defect resolution, client deployments), and a growing portfolio of production AI agents and automation systems. Certified in cybersecurity through Google Cloud and Cisco, a security-conscious foundation for the private, local-first AI I build. I build things that run in production, not demos.

Available for remote consulting engagements and agentic / AI engineering roles, worldwide.