AI Process Automation Portfolio

Implemented automation and integration projects focused on APIs, webhooks, workflow orchestration, persistence, auditability, reliability patterns and bounded AI-assisted processing.

10-second summary: the strongest modules are a signed webhook integration hub and a workflow automation MVP. Both are local portfolio systems with explicit failure handling and scope boundaries; mocked external systems and illustrative ROI estimates are not presented as production evidence.

Recruiter Snapshot

Module	What it demonstrates	Core technologies	Status
01-rest-api-integration-hub/	Signed webhook intake, transformation, idempotency, replay protection, mock system integration, audit and DLQ-style handling	FastAPI, PostgreSQL, Docker Compose, n8n, HMAC	Implemented local portfolio project
02-ai-workflow-automation-engine/	Operational intake, structured AI-assisted extraction, routing, persistence and workflow metrics	FastAPI, PostgreSQL, n8n, Docker Compose, LLM adapter	Implemented MVP
00-rest-api-basics/	Earlier REST/API exercises	Python / HTTP basics	Learning sandbox

01 — REST API Integration Hub

A local integration hub for signed webhook intake and downstream automation.

Signed webhook
      ↓
FastAPI validation
      ↓
HMAC / timestamp checks
      ↓
Idempotency + replay protection
      ↓
Transformation
      ↓
PostgreSQL / audit / DLQ-style handling
      ↓
n8n orchestration
      ↓
Mock CRM + mock Jira/Confluence targets

Implemented evidence

• FastAPI webhook/API layer
• HMAC-signed requests using timestamp and signature headers
• payload validation and transformation
• idempotency and replay-protection logic
• audit logging
• Dead Letter Queue / error-path handling
• KPI aggregation
• n8n orchestration
• Docker Compose local environment
• smoke tests for happy and error paths

Boundary: downstream business systems are simulated. The project demonstrates integration mechanics and safeguards, not a real enterprise deployment.

→ Open project

---

02 — AI Workflow Automation Engine

A webhook-driven MVP that turns unstructured operational intake into structured workflow state.

Operational intake
      ↓
FastAPI
      ↓
Structured extraction
      ↓
Routing logic
      ↓
PostgreSQL persistence
      ↓
n8n branching
      ↓
Mock Jira / Mock Slack targets

Implemented evidence

• POST /intake
• GET /health
• GET /workflow/{id}
• GET /metrics/summary
• mock Jira and Slack endpoints
• structured extraction of summary, action items, priority and route
• deterministic mock mode plus OpenAI-compatible adapter mode
• persisted workflow state, outputs, events, errors and metrics
• n8n workflow export
• pytest-based test execution

Boundary: the automation-impact metric uses configurable manual/automated time assumptions. It is an illustrative model, not validated business ROI.

→ Open project

Technical Stack by System Role

Role	Evidence
Connect / Intake	REST APIs and webhooks
Buffer / Workflow state	persisted workflow/event records; no dedicated message broker is claimed in these modules
Processing	validation, transformation, routing and bounded AI extraction
Storage	PostgreSQL
Orchestration	n8n
Security	HMAC verification, timestamps, replay protection, idempotency
Reliability	validation, explicit error paths, DLQ-style handling, smoke tests
Auditability	audit logs plus persisted workflow/event state
AI boundary	adapter-based structured extraction; AI is workflow assistance, not autonomous authority
Infrastructure	Docker Compose

Engineering Lessons and Failure Modes

The useful engineering work in these projects is not simply calling an API or an LLM.

• Webhook authenticity must be checked before business processing. Signature validation and timestamp handling belong at the trust boundary.
• Retries create duplicate risk. Idempotency and replay protection are necessary when the same event may arrive more than once.
• Failure needs a destination. Validation or downstream errors should remain inspectable instead of disappearing into logs only.
• Workflow state must be persisted. A successful HTTP response is not enough to explain what happened later.
• External integration should be separated from internal logic. Mock adapters make the portfolio implementation reproducible while keeping the boundary explicit.
• Probabilistic AI output needs deterministic surrounding contracts. Structured extraction is bounded by routing and persistence logic rather than being treated as a complete autonomous workflow.
• Estimated time savings are assumptions until measured in a real process. The repository keeps that distinction explicit.

Scope and Limitations

This repository does not claim:

• production authentication or enterprise IAM
• real Jira, Slack, CRM or Confluence deployment
• enterprise hosting or production observability
• multi-agent autonomy
• validated financial ROI
• production operational ownership

Suggested Review Path

1. 01-rest-api-integration-hub/ — strongest evidence for integration security, reliability and failure handling
2. 02-ai-workflow-automation-engine/ — structured AI-assisted workflow state and persistence
3. 00-rest-api-basics/ — earlier exercises only

TL;DR

This repository demonstrates how I approach automation as an engineering system: trust boundary → validation → idempotency/replay handling → transformation → persistence → orchestration → explicit failure paths → auditability. AI is added only inside that controlled workflow rather than used as a substitute for system design.