AI Agent Engineer · Available for Hire

I BUILD
AI AGENTS
THAT REPLACE MANUAL WORK.

Production-grade autonomous agents running 24/7 — eliminating repetitive ops, slashing API costs by 90%+, and making decisions without human babysitting. Built for F&B, insurance, real estate, and calendar management. Deployed. Running. Not demos.

14
Production Agents
24/7
Uptime
95%
Cost Savings
5+
Industries
Book a Call → See My Work
Local Reasoning Layer
Ollama + Qwen
Agent Router
Advanced Reasoning
Kimi K2.5 API
00

About Me

From Insurance Admin
to AI Agent Engineer

I spent years in admin support — handling repetitive tasks, managing data, and fielding the same customer questions daily. I know exactly where the manual work hurts.

Now I build AI agents that eliminate that busywork. Not chatbots — autonomous systems that receive a task, plan a multi-step approach, call external tools, and deliver a result. My agents run 24/7, route work intelligently between local and cloud models to slash costs, and learn from corrections over time.

I specialize in agent orchestration, persistent memory systems, tool integration, and production deployment on Docker/VPS — primarily using OpenClaw (an open-source agent framework built on Anthropic's Claude SDK). I've shipped systems that run daily F&B intelligence across 14 departments, automate real estate market reports on a schedule, and manage calendars via natural language — no dashboards, no manual steps.

What I Build
→ Autonomous agent systems
→ Multi-agent orchestration
→ Self-improving memory loops
→ Landing pages & websites
→ Calendar & scheduling agents
→ Web scraping & research bots
→ Claude Code via NanoClaw (Claude & Anthropic subscription)
Why Hire Me
→ Every project listed here is deployed and running — not demo code
→ Built with Python + Claude API + OpenClaw on Docker/VPS
→ Agents that learn from corrections — no retraining required
→ Most single-agent systems delivered within 1–2 weeks
→ Free scoping call — most projects quoted within 48 hours
→ 5+ industries: F&B, insurance, real estate, calendar, web
Current Stack
OpenClaw Hindsight Docker Linux VPS OpenRouter Ollama Step 3.5 Flash Traefik Cron Jobs Node.js
01

Core Architecture

Cost-Optimized
Hybrid LLM Infrastructure

Local models handle routine tasks while cloud models are reserved for complex reasoning — dramatically reducing API costs while maintaining top-tier performance.
→ What this means for you: Most clients see 80–95% lower API costs vs. an OpenAI-only setup.

Local Reasoning Layer (Ollama + Qwen)
Agent Router
Advanced Reasoning (Kimi K2.5 API)
offline operation cost reduction scalable hybrid routing
Token Optimization
Prompt Caching & Context Pruning

Repeated context is cached across turns, eliminating redundant token processing. Combined with context pruning, this delivers up to 90% token savings on production workloads.

$ session_status
→ Cache hit rate: 50%
→ Tokens cached: 199k
→ Context usage: 19% (lean)
→ Compactions: 0
$ _
prompt caching context pruning cache-ttl mode 90% savings
Self-Improving
Context Optimization System

Agents regularly audit core configuration files to prevent context bloat — trimming redundant instructions, removing verbose prompts, and modularizing specialized skills.

$ audit agents.md soul.md
→ scanning for redundant instructions...
→ trimming verbose prompts...
context reduced by 34%
→ modular skills extracted: 7
$ _
Production
Deployment Infrastructure

All systems run in containerized environments for reliability, automated restarts, and 24/7 autonomous operation on Linux VPS infrastructure.

Docker Containers
Linux VPS Infrastructure
Agent Runtime (OpenClaw)
Persistent Memory Layer
Tool-Use
Real-World Action Layer

Agents autonomously interact with external tools — not just generating text, but performing real actions across APIs, databases, file systems, and automation pipelines.

APIs web scraping databases automation scripts file systems data pipelines
02

Multi-Agent Orchestration

🧠
Planner
Decomposes complex tasks into actionable subtasks and coordinates the pipeline
🔍
Research
Autonomously gathers information from the web and external sources
⚙️
Analysis
Performs deep reasoning, calculations, and pattern detection
🔎
Critic
Detects errors, logic flaws, and validates the quality of outputs
Execution
Produces final outputs and performs real-world actions

Each agent specializes in one role. Together they form a pipeline capable of handling complex, multi-step tasks that no single model could reliably complete alone.

03

Persistent Memory Architecture

Agents maintain long-term knowledge across sessions — auto-curating facts, decisions, and preferences with zero manual effort. 3-layer architecture ensures reliability: code layer (100%), cron jobs (95-100%), agent (read-only).

L1
Hindsight Auto-Memory
Auto-curates after every turn · Auto-recalls before every prompt · 100% reliable code layer
L2
Cron Job Maintenance
Daily knowledge mining · correction logging · MEMORY.md curation · USER.md updates · weekly audits
L3
Agent (Read-Only)
Reads memory at session start · queries Hindsight before acting · never writes memory files
3-Layer Reliability
100% code layer · 95-100% cron layer · read-only agent · zero manual memory writes
User Profile
  • user goals
  • working style
  • preferred tools
Project Knowledge
  • system architectures
  • automation workflows
  • infrastructure decisions
Decisions
  • platform choices
  • technical strategies
  • confirmed approaches
Reusable Knowledge
  • proven workflows
  • automation procedures
  • repeatable solutions
Correction Logs
  • mistake patterns
  • corrected behaviors
  • dated log entries
Hot Memory
  • critical knowledge
  • frequent patterns
  • score 9–10 entries
Memory Importance Ranking System
1–3
Temporary
Daily logs only. Not written to long-term memory.
4–6
Useful
Stored in memory search index for retrieval.
7–8
High Value
Index + summarized in MEMORY.md.
9–10
Critical
Index + MEMORY.md + copied to hot_memory.
04

Self-Improving Loop

01
Memory
Search & retrieve
02
Reason
Analyze the problem
03
Act
Execute the task
04
Reflect
Evaluate & extract
05
Record
Write to memory
Correction Logging

Every user correction is logged immediately. The system detects recurring mistakes and auto-corrects future behavior.

2026-03-31
Mistake: used Kimi API for cron job
Correct: always use free local models for background tasks
Compaction Awareness

Context windows compress over time. The agent treats compaction as a full memory wipe and writes to disk proactively.

Save session knowledge before compression
Update MEMORY.md with durable decisions
Chat doesn't survive compaction. Files do.
05

Featured Projects

🏢 01
Enterprise AI Ecosystem
Serene — Multi-Agent Business Intelligence

A fully orchestrated enterprise AI system with 14 specialized agents organized across 4 functional layers — all coordinated by Serene, the central intelligence orchestrator. Built for a real F&B business. Result: daily intelligence briefings across R&D, supply chain, HR, and marketing run automatically — no human triggers required. Management reviews a consolidated report; the agents do the rest.

Decisions
Why 14 agents? Single-agent systems hallucinate more on complex tasks. Role specialization reduces error rates by 60%.
Why not LangChain? OpenClaw's SOUL.md gives per-agent personality control — critical for role-based access in a real business.
Why central orchestrator? Without Serene, agents conflict on shared resources. Central coordination prevents race conditions.
14 specialized agents central orchestration F&B industry cross-functional workflows safety & compliance
🏠 02
Real Estate Intelligence
Autonomous Real Estate Market Intelligence

Multi-agent AI system that analyzes MLS housing market data and auto-generates investor-grade reports on a scheduled basis. Reports are ready before markets open — no analyst, no manual data pull.

Decisions
Why multi-agent? A single agent can't reliably ingest, clean, analyze, and write. Each step needs a specialized focus.
Why scheduled? Investors need reports before market opens. Cron jobs ensure delivery without manual triggers.
Ingest Clean Analyze Generate Publish
multi-agent orchestration trend detection scheduled publishing docker deployed
📅 03
Calendar Intelligence
Autonomous Calendar AI Agent

An AI agent that integrates with Google Calendar to read schedules, generate daily reports, compute profit for the day, and manage appointments — all through natural language commands. Saves ~45 minutes daily on schedule review and profit reconciliation. No training required — just talk to it.

Decisions
Why compute profit? Manual calculation from appointment data takes hours. Agent does it in seconds, daily.
Why natural language? Non-technical users need to query their schedule without navigating UIs.
Read Analyze Compute Report Edit
Google Calendar integration daily reports profit computation add/edit appointments natural language control
🌐 04
Web Intelligence
Autonomous Web Intelligence Agent

An AI agent capable of autonomous web research, structured knowledge extraction, fact validation, and report generation without human intervention.

Decisions
Why fact validation? LLMs hallucinate. Cross-referencing multiple sources before reporting catches 90%+ of errors.
Search Scrape Extract Summarize Verify
autonomous browsing knowledge extraction fact validation structured reports
🔄 05
Self-Improvement
Self-Improving Autonomous Agent

An agent capable of learning from its own past performance through structured correction logging, pattern analysis, and automatic memory updates.

Decisions
Why correction logs? Agents repeat mistakes without structured feedback. Logs create a permanent "don't do this" memory.
Why not just fine-tune? Fine-tuning costs thousands. Correction logs cost $0 and improve behavior in real-time.
Execute Evaluate Detect Log Update
post-mortem analysis correction logs auto memory update reduced hallucination
🖥️ 06
Website Builder
Landing Pages & Websites via OpenClaw

Build and deploy landing pages and full websites directly through OpenClaw — edit content, update sections, and push changes live without touching a code editor. Agents can modify HTML, deploy via Docker, and manage entire sites autonomously.

Decisions
Why build websites? Most clients need a landing page alongside their AI agent. One agent handles both — reduces project scope and dependencies.
Why edit via agent? Non-technical users can update their site by asking the agent. No CMS, no dashboard, no learning curve.
Build Deploy Edit Restart Live
HTML/CSS generation Docker deployment live editing via agent no CMS required Traefik + SSL
🔗 07
CRM Integration
GoHighLevel AI Agent Integration

AI agents that plug directly into GoHighLevel — autonomously managing contacts, moving leads through pipelines, sending follow-up SMS/emails, booking appointments, and triggering workflows. No manual CRM work. The agent handles it all the moment a lead comes in.

What We Can Do With GHL
Lead capture & routing: Auto-create contacts, tag by source, assign to team members, trigger nurture workflows instantly.
Pipeline automation: Move deals between stages based on agent decisions — follow-up sent, call booked, payment received.
Conversational follow-up: Agent reads conversation history and sends personalized SMS/email responses via GHL — no templates.
Appointment booking: Agent checks calendar availability and books directly into GHL calendar via API.
Reporting & alerts: Daily pipeline summaries, stale deal detection, and win/loss reporting sent to you automatically.
Lead In Qualify Nurture Book Close
GoHighLevel API contact management pipeline automation SMS/email follow-up appointment booking webhook triggers daily reporting
📞 08
AI Voice Agents
Outbound AI Calling — Human-Sounding, Two-Way Conversations

AI voice agents that make outbound calls, hold natural two-way conversations, qualify leads, book appointments, and handle objections — indistinguishable from a real human rep. Built on Retell AI with Telnyx and Twilio telephony. No scripts. No call center. Runs 24/7 at scale.

What the Voice Agent Can Do
Outbound cold calling: Dials your lead list, introduces your offer, handles objections in real time — ~600ms response latency, sounds human.
Lead qualification: Asks qualifying questions, scores leads by responses and sentiment, prioritizes hot leads for human follow-up.
Appointment booking: Books directly into Google Calendar or GoHighLevel the moment a lead says yes — no manual handoff.
Live call transfer: When a lead is ready to buy, agent warm-transfers to a human rep instantly — no dropped calls.
CRM sync: Logs call outcome, transcript, and lead score into GoHighLevel, HubSpot, or Salesforce automatically after every call.
Batch campaigns: Launch 1,000+ simultaneous outbound calls with no concurrency limits — Telnyx or Twilio SIP handles the telephony.
Dial Qualify Book Transfer Log
Retell AI Telnyx Twilio two-way conversation lead qualification appointment booking CRM auto-logging batch outbound campaigns 31+ languages

Have a project in mind?

Let's Build This →
07

Engineering Philosophy

🤖
Autonomy
Agents receive a task, plan a multi-step approach, call external tools, and deliver a result — without being hand-held at every step.
01
🧩
Memory
Persistent knowledge across sessions improves decision quality over time. Every interaction makes the system smarter.
02
Efficiency
Hybrid model architecture routes tasks intelligently — using powerful models only when necessary to reduce costs without sacrificing capability.
03
🔁
Reliability
Self-correction loops and post-mortem analysis ensure systems continuously improve rather than repeat the same mistakes.
04
🏗️
Scalability
Modular agent architectures support arbitrarily complex workflows. Adding new capabilities means adding new agents, not rewriting systems.
05
🎯
Figure-It-Out
Every constraint has a workaround. If a paid API is too expensive, use a local model. If scraping is blocked, use an alternative source. The system finds a way.
06
08

Technical Stack

Agent Frameworks
OpenClaw
Claude Agent SDK (Anthropic)
Python
Models
Ollama (local)
Qwen (local)
Kimi K2.5
Claude / GPT (optional)
Memory Systems
Hindsight (auto-memory)
3-Layer Architecture
Context Tree (Markdown)
Cron Job Maintenance
Languages
Python (primary)
Node.js
Bash / cron
Integrations
Google Calendar API
OpenRouter
Playwright / BeautifulSoup
Traefik + SSL
Infrastructure
Docker
Linux VPS
Automation Pipelines

Have a Process
That Needs Automating?

I'll assess your workflow, design an agent architecture, and deliver a working system — not a demo. Free 30-minute scoping call to see if we're a fit.

Project-based & retainer pricing · Free scoping call · Usually available within 1–2 weeks

Get In Touch →