JAMES MONTOYA
AI Engineer — Document AI, RAG Systems & Agentic Pipelines
Sydney, Australia · Australian Citizen | james@ethicalaispecialists.com.au | linkedin.com/in/james-montoya | jamonhin.github.io
Professional Summary
I build production AI systems that ship. Sydney-based AI engineer with twenty years across enterprise data engineering and, for the last three, LLM-powered automation and production AI. Twelve of those years were spent building ETL pipelines, data models and reporting platforms in banking and telecommunications — which is why I treat an AI system as a data system: ingestion, transformation, storage, retrieval, presentation. Core stack: Python, n8n (self-hosted), ArangoDB, PostgreSQL, AWS, Docker, and local LLM inference where data cannot leave the building. Track record of cutting manual processes by up to 95% and taking systems from idea to production in days rather than months, in regulated settings: financial services, defence recruitment, not-for-profit, banking and telecommunications.
Show, Don't Tell — Featured Work
DocTriage — AI Document-Validation Pipeline (no-interest microloan assessments)
  • Reads and validates an applicant's full document pack — identity, income, banking, tenancy, supplier quotes — against eligibility rules encoded as versioned, auditable decision tables, each citing its source in the lending handbook
  • Two extraction paths behind one contract: a digit-perfect native-PDF extractor (no GPU) and a GPU vision-OCR service for scans and images
  • Grounded extraction — every value is verified against the document's own text; anything low-confidence escalates to a human instead of flowing silently into a decision
  • Privacy by design: inference runs locally, real applicant data never leaves the environment, development runs entirely on synthetic applicants. The system never auto-approves
  • Explainer: youtu.be/o7USxW7kreY  ·  Live demo: youtu.be/Z-52Q6Odrcw
Voice of the Customer Analytics Platform (built as a weekend POC)
  • End-to-end pipeline: real insurance call transcripts → LLM extraction (Gemini 2.5 Flash) → ArangoDB storage → live webhook dashboard
  • 16-field structured extraction: calibrated sentiment scoring, sentiment journey tracking, churn risk, topic classification, agent performance
  • Benchmarked GPT-4.1 Mini vs Gemini 2.5 Flash on identical transcripts — selected Flash for superior nuanced sentiment detection
  • Walkthrough: youtu.be/aUAvT4RlgSw
  • Full project portfolio: jamonhin.github.io
Professional Experience
AI Consultant — Independent Practice
Nov 2023 — Present
Ethical AI Specialists
Sydney, Australia

Independent AI consultancy delivering production systems for three concurrent clients across not-for-profit, defence recruitment and financial services.

Australian Not-for-Profit · Nov 2024 — Present · Sydney
  • DocTriage (since May 2026) — AI document-validation pipeline for no-interest microloan assessments. Classification and field extraction across eight document categories, eligibility rules encoded as versioned auditable decision tables, grounded extraction with low-confidence escalation to a human
  • Privacy-by-design architecture: local LLM inference, synthetic applicants in development, no real applicant data leaving the environment. The system never auto-approves — a caseworker makes every final decision
  • Created and delivered an AI training video course for not-for-profit teams, covering safe adoption, data governance and practical tooling
  • Automated the end-to-end lead data pipeline for Meta digital acquisition campaigns
  • Stack: Python, Docker, vLLM, vision OCR, decision-table rules engine, n8n, SQLite, CRM integration
Defence & Government Recruitment Firm · Jun 2025 — Present · Canberra (Remote)
  • Built multi-stage AI recruitment pipeline: CV ingestion, structured extraction (skills, security clearances, capability domains), graph-based storage in ArangoDB
  • Designed multi-agent architecture (Router → Query Builder → Response Formatter → Validator) orchestrated through n8n, surfaced to recruiters as a conversational agent inside Microsoft Teams
  • Extended the platform into executive reporting: n8n ingestion of HubSpot contacts and deals into an ArangoDB reporting warehouse, a read-only FastAPI service exposing 16 metric endpoints, and two Grafana dashboards provisioned as code and deployed on AWS EC2
  • HubSpot integration processing ~1,759 contacts with AGSVA clearance and skills data; automated resume processing cutting manual data entry by 95%; dual-layer skill extraction preventing domain mismatches with 90%+ accuracy
eThink Solutions · Nov 2025 — Present · Brisbane (Remote)
  • RAG system adding document intelligence to a custom-built mortgage broker CRM — 150+ loan documents at 95%+ accuracy using Claude as the LLM layer, cutting manual lookup time by 80%
  • Designed an ArangoDB multi-model schema combining document store, graph and vector search for classification, chunk-level tagging and relationship mapping between banks, processes and compliance requirements
  • Sage — multi-tenant voice-AI platform where elderly people record life stories by talking to a conversational AI and choose, story by story, what their family can hear. The browser opens a WebRTC connection straight to the OpenAI Realtime API using short-lived server-minted tokens, so the API key never reaches the client; consent is enforced in Postgres row-level security rather than in the UI. Eleven languages with cached translations and semantic search across memories. Demo: youtu.be/xZAFn2mvJYw
  • AwareIM XML Generator — a Claude skill turning a Markdown application spec into import-safe AwareIM BSV XML, gated by a 19-check pre-flight validator so a business space the platform would reject never ships
  • Stack: n8n, ArangoDB, Claude API, AWS (EC2, S3), Docker, Traefik, Python, Next.js, TypeScript, Supabase (Postgres, RLS, pgvector), OpenAI Realtime API
Earlier independent work · Nov 2023 — Nov 2024
  • Production-ready RAG chatbot with PostgreSQL/pgvector, Next.js 15, React and Redis; Jobseeker multi-agent system using LangChain, CrewAI and OpenAI
  • Docker-based ML environments, automated ETL with Apache Airflow, and predictive models for nonprofit donor behaviour using clustering
  • Completed Master of Software Engineering (AI Specialisation), Torrens University
Relocation to Australia & Postgraduate Study
2017 — 2023
Sydney, Australia
  • Relocated from Colombia to Australia; now an Australian citizen
  • Completed a Master of Software Engineering (AI Specialisation) at Torrens University Australia, pivoting from enterprise data engineering into applied AI
Data Engineer & Business Intelligence Developer
2014 - 2017
Tigo (Telecommunications)
Colombia
  • Built a mobile KPI dashboard app for executives: Apache PhoneGap front end backed by a PHP services API over SQL Server, including interface design
  • Created ETL packages in SSIS for seamless multi-source data integration
  • Automated reporting systems using SQL Server, reducing manual effort by 80%
  • Implemented data scraping and processing solutions with Python
  • Designed real-time data visualization systems using QlikView for executive decision-making
  • Stack: SQL Server, SSIS, PHP, Python, QlikView, Apache PhoneGap, JavaScript, VBA
Analyst Developer & Database Specialist
2006 - 2014
Grupo Bancolombia (Financial Services)
Colombia
  • Led treasury database reporting migration in a regulated banking environment, ensuring audit compliance and seamless transition
  • Developed Java applications for Murex financial system integration and data processing
  • Implemented reporting solutions using Datamart and SAP Business Objects
  • Managed Murex system transition, optimizing financial data processes and system integration
  • Created VBA macros automating spreadsheet-to-production migrations
  • Leveraged Oracle Database and Business Objects for stakeholder reporting
  • Stack: Oracle, SQL, Java (Murex), VBA, SSIS, SAP Business Objects, Datamart, Murex
Independent Projects
Jobseeker Multi-Agent System
Jan 2025
Python, LangChain, CrewAI, OpenAI, React
Developed intelligent job application system analyzing postings and generating tailored cover letters and optimized resumes using multi-agent AI architecture.
Cleaning Company Management System
May 2025
PostgreSQL, N8N, OpenAI GPT, Docker
Multi-agent system automating operations and providing actionable insights for cleaning service providers with 30+ interrelated database tables.
ML-Powered Donor Analysis System
2023-2024
Docker, Python, Apache Airflow, Scikit-Learn, KMeans
Comprehensive system analyzing donor behavior and predicting contributions for nonprofit organizations using containerized ML environments.
Technical Skills
AI & LLM Engineering
Claude (Anthropic API), OpenAI, Gemini, Amazon Bedrock, Ollama, local inference (vLLM), prompt engineering, multi-agent orchestration, RAG systems, document AI & OCR, vector search
Frameworks & Orchestration
n8n (self-hosted, production-grade), LangChain, LlamaIndex, CrewAI, FastAPI, decision-table rules engines
Programming
Python, TypeScript, JavaScript, SQL, AQL, Node.js, React, Next.js, Java, PHP
Databases
ArangoDB (document, graph, vector search), PostgreSQL, pgvector, Supabase, Neo4j, SQL Server, Oracle, MySQL
Cloud & Infrastructure
AWS (EC2, S3, Lambda, Bedrock, OpenSearch, API Gateway), Docker, Traefik, Nginx, Let's Encrypt, Azure (Teams, SharePoint, OneDrive APIs), Google Cloud, serverless architectures
Data Engineering
ETL pipelines (SSIS, Airflow, n8n), data modelling, multi-source integration, MLOps, Datamart, reporting warehouses, Grafana
Integration & APIs
HubSpot, Salesforce, Azure, Google APIs, webhooks, any system with API connectivity
Also experienced with
Power BI, Tableau, QlikView, VBA, Scikit-Learn, Chart.js — from twelve years in enterprise BI and data engineering
Education
Master of Software Engineering (AI Specialization)
2021 - 2023
Torrens University Australia
Sydney, Australia
Advanced studies in machine learning, artificial intelligence, and software development with focus on practical AI applications.
Diploma of Software Development
2018 - 2020
Warwick Institute of Australia
Sydney, Australia
Two-year diploma covering software development lifecycle, programming and database design, completed after relocating to Australia.
Certificate III in English
2018
TAFE NSW
Sydney, Australia
English language study on arrival in Australia.
Specialization in Software Development
2013
Universidad de MedellĂ­n
Colombia
Software development lifecycle, Scrum methodologies, requirements elicitation, and software engineering best practices.
Bachelor of Computer Science
2009
Politécnico Jaime Isaza Cadavid
Colombia
Foundation in computer science, programming, databases, and software development principles.
Key Achievements
  • Designed an auditable document-validation pipeline whose eligibility rules live in versioned decision tables, each citing its source policy — so every automated outcome can be traced back to the rule that produced it
  • Shipped privacy-by-design AI: local inference, synthetic data in development, and no real applicant information leaving the client environment
  • Built a reporting warehouse and two provisioned-as-code Grafana dashboards over a HubSpot → ArangoDB pipeline, deployed on AWS EC2
  • Reduced manual document lookup time by 80% and manual data entry by 95% through AI-powered automation
  • Achieved 95%+ accuracy in intelligent document processing across 150+ documents
  • Built the Voice of the Customer POC in days — end-to-end pipeline on real insurance transcripts with a live dashboard
  • Designed multi-agent AI architectures for production enterprise environments
  • Successfully migrated complex financial database systems at enterprise scale in regulated banking
  • Delivered AI training programmes for not-for-profit organisations
Core Competencies
  • LLM Pipeline Design & Deployment
  • Multi-Agent AI Architecture
  • RAG System Implementation
  • Workflow Orchestration (n8n)
  • Graph Database Design (ArangoDB)
  • Enterprise Data Integration
  • Business Process Automation
  • Cross-functional Stakeholder Communication