Brian Oktavec
Projects

Systems I've built, and what they had to survive.

Four projects spanning platform engineering, security, and AI — from a full-stack compliance SaaS to a local-first secrets manager.

Atlas

AI / LLMPERSONAL
self-hosted RAG knowledge assistant

A privacy-focused Retrieval-Augmented Generation platform that turns private documents and notes into a searchable AI assistant. An ingestion pipeline extracts and normalizes metadata, chunks content, generates embeddings, and stores them in Qdrant; queries are matched via semantic search and answered by a locally hosted LLM, with intelligent routing to external sources when more context is needed.

n8nQdrantVector EmbeddingsLocal LLMsDocker
Case study →

Inspectra

PLATFORMWORK
full-stack inspections/compliance SaaS

A customer portal, staff back office, and offline-capable inspector PWA on an AWS/Kubernetes backend. Built a shared JWT/OIDC auth package with RBAC, a Puppeteer + React PDF reporting framework, and an offline-first PWA with service worker sync and conflict resolution.

AWSKubernetesPostgreSQLTerraformNext.js
Case study →

Keyveil

SECURITYPERSONAL
local-first secrets management for developers & AI agents

A macOS secrets management ecosystem built around Apple Keychain: a native app, a CLI, an MCP bridge that lets AI agents reference secrets without seeing raw values, and a Raycast extension — designed around least-privilege, local-first access.

SwiftApple KeychainMCPRaycastCLI
Case study →

MailForge

AI / LLMPERSONAL
AI-powered email classification & automation engine

A self-hosted email classification and automation system that integrates with Fastmail via JMAP. n8n orchestrates ingestion and mailbox actions, while a locally hosted LLM (Qwen) determines category, priority, and destination folder for each message — combining deterministic rules with LLM-based classification and tracking processed messages to prevent duplicate handling.

n8nFastmail JMAPLocal LLMsQwenWorkflow Automation
Case study →