Customer Support Automation
Multi-channel intake, intent classification and grounded answers, with escalation paths when confidence or sentiment drops.
AI Automation Engineer
Bambang Saputra Jaya
I build production AI systems that replace manual business work — not demos.
Business workflow automation with n8n, Python/FastAPI, LLMs, RAG, APIs, and reliable control layers.
- 6 years enterprise software engineering & system integration
- Production AI automation
- n8n + Python/FastAPI + PostgreSQL/pgvector + LLM APIs
What I Build
Automation that removes recurring manual work
Focused systems that run in production, with the boring reliability parts included.
Invoice & Document Processing
Turn PDFs and unstructured documents into validated, schema-conformant records persisted in your database.
AI Agents & RAG
Retrieval-grounded assistants over your own knowledge base, with policy enforcement instead of free-form guessing.
Workflow Automation
Repetitive back-office processes rebuilt as deterministic workflows with retries, idempotency and audit trails.
API & System Integration
Connect CRMs, databases, messaging platforms and internal services into one reliable control layer.
Flagship Systems
Four systems, end to end
Each one is a working repository — architecture, workflow logic and control layers included.
System previewAI Customer Support Automation Platform
Problem
Support teams handle the same high-volume requests manually across multiple channels, with inconsistent answers and no reliable escalation.
Solution
A 77-node n8n workflow with PostgreSQL + pgvector and OpenAI: multi-channel intake, 15-intent classification and RAG-grounded responses under deterministic policy enforcement.
Key engineering features
- Multi-channel intake and 15-intent classification
- RAG over PostgreSQL + pgvector
- Deterministic policy enforcement
- Sentiment-driven escalation and human fallback
- Idempotency and structured-output validation
- Observability across the workflow
n8nPostgreSQLpgvectorOpenAIRAG
View on GitHub
System previewAI Invoice Processing Pipeline
Problem
Invoice data is re-typed by hand from PDFs into internal systems, which is slow and error-prone at volume.
Solution
A self-hosted, reproducible pipeline: PDF ingestion, text extraction with an OCR-oriented workflow, Ollama/local LLM extraction and required-schema JSON validation before persistence.
Key engineering features
- PDF ingestion with text extraction / OCR-oriented workflow
- Ollama / local LLM extraction
- Required-schema JSON validation
- PostgreSQL persistence and HTTP response
- Explicit error handling
- Self-hosted and reproducible
n8nOllamaPostgreSQLJSON SchemaSelf-hosted
View on GitHub
System previewWhatsApp AI Customer Support
Problem
Customers ask the same questions on WhatsApp all day, and staff answer them one message at a time.
Solution
WhatsApp customer support automation using n8n and Evolution API, with a local Ollama model, containerised services and a PHP REST API backed by MySQL.
Key engineering features
- WhatsApp intake via Evolution API
- Ollama-powered response generation
- FAQ handling
- PHP REST API and MySQL backend
- Dockerised deployment
- Workflow automation end to end
n8nEvolution APIOllamaDockerMySQLPHP
View on GitHub
System previewAI Recruitment Automation
Problem
Recruiting pipelines lose time on manual posting, applicant collection and first-pass CV screening.
Solution
An end-to-end recruitment workflow covering job posting, applicant intake, AI screening, scoring and routing to the right next step.
Key engineering features
- Job posting automation
- Applicant intake
- AI screening
- Candidate scoring
- Routing to next steps
- End-to-end workflow orchestration
n8nLLM APIsAutomationIntegrations
View on GitHubEngineering Principles
What makes an automation safe to run unattended
Validation
Model output is validated against a required schema before anything is written or sent.
Idempotency
Retries and duplicate events do not create duplicate side effects.
Guardrails
Deterministic policy rules constrain what the system is allowed to answer or do.
Fallback & Human Escalation
Low confidence or negative sentiment routes the case to a person instead of guessing.
Auditability
Every decision and record is traceable back to its input and the step that produced it.
Observability
Runs, failures and bottlenecks are visible so issues are found before users report them.
About
Enterprise engineering background, AI automation focus
I spent 6 years in enterprise software engineering and system integration at PT Astra Digital Internasional, working on systems where correctness, integration and reliability matter more than novelty.
Today my focus is AI automation for SMEs, SaaS companies, agencies, and remote-first teams: identifying the repetitive workflow, designing the control layer around the model, and shipping something that keeps running after launch.
I work with n8n, Python/FastAPI, PostgreSQL and pgvector, LLM APIs and local models, and the APIs and integrations that connect them to the tools a business already uses.
Have a repetitive workflow that should be automated?
Let's identify the workflow, estimate the automation opportunity, and define a practical implementation path.
Book a 30-min automation audit