Justin Blair
14 months from “what’s Docker?” to shipping three production systems. Mediastack serves close friends and family seeking alternatives to extractive big tech. Stars enters invite-only beta in the coming weeks. 547 documented bug fixes. Multi-agent workflow simulates team review. All built in spare time while working full-time. No CS degree. No team. Remote.
What I’m Looking For
Role
Backend engineer, infrastructure engineer, platform engineer, AI infrastructure, or security operations. Currently building an AI-agent-driven frontend suite for the Stars Engine’s 100+ pages — my agents write the UI.
Location
Remote. I operate a multi-host WireGuard mesh across two states — distributed work is my native mode.
Company Stage
Startup to growth-stage. I thrive where broad skills matter more than narrow specialization and shipping velocity is valued.
Compensation
Open to discussion based on role and responsibility. I bring 14 months of demonstrated output — evaluate the portfolio, not the years.
Not Interested In
Ad-tech, surveillance capitalism, crypto/web3, gambling, or anything that extracts value from users without their informed consent. I build tools people control — not tools that control people.
Philosophy
Self-hosted infrastructure. Data sovereignty. Source-cited outputs. Deterministic computation. Privacy by design. AGPL where it protects users. If your company shares these values, we should talk.
Demonstrated Capabilities
Backend Systems
53-endpoint REST API with JWT auth, CSRF middleware, rate limiting, and structured error handling across 17 route modules. Consistent patterns throughout.
Data Architecture
Bronze/Silver/Gold medallion pipeline: content-addressed raw storage (SHA-256, 152K charts), PostgreSQL serving layer, DuckDB/Parquet analytics across 10K charts.
Knowledge Representation
5,070-node concept graph with 3,550 edges, 3 edge types, 5 source tiers, and 22 hyperedges. 316 doctrinally attested edges with source citations. 14 regression tests.
Distributed Systems
Redis Streams consumer groups with PEL crash recovery. LLM-based task classification with confidence gating. HNSW semantic routing. Thompson sampling bandit. Circuit breaker with fallback.
Security Architecture
mTLS on message bus. HMAC-signed inter-agent messages with replay protection. Per-agent ACL. Time-boxed JIT permission tokens. PII scrubbing. Key rotation with fallback.
Concurrency Control
Multi-agent file locking with lock reaper for crashed agent recovery. Trust-scoped access with 4 ATF levels (Intern/Junior/Senior/Principal) gating filesystem and network access.
Uniform Interfaces (ABI)
32 resource handlers exposing Docker containers, hardware devices, and APIs through the same 6 verbs. A database and a light bulb share the same interface. On-the-fly handler generation.
Adversarial Verification
P0/P1 task results auto-reviewed by a different AI model family. Claude output gets Gemini review. Caught 3 real bugs in one day that same-model review missed.
Self-Healing Infrastructure
Agents detect failures, trace dependencies, restart root causes, and verify repairs — across containers, downloads, and hardware — without playbooks, just interface contracts.
Autonomous Cybersecurity
Suricata detects, Wazuh correlates, CrowdSec checks threat intel, agent decides: block IP (fail2ban), isolate container (Docker proxy), capture forensic snapshot. No human in the loop.
Security Auditing
Audited own codebase: found SQL injection, XSS, path traversal, broken error attribution, and silent exception swallowing. Fixed 18 of 30 findings.
Infrastructure as Code
11 layered compose files. 188 Docker/K8s/YAML configs. 101 shell scripts. Traefik reverse proxy. nginx static sites. WireGuard mesh. tmux agent dashboard.
Testing Discipline
983 pytest tests. 14 graph content regression tests. Public accuracy benchmark with verified positions. Build-time invariants that gate graph quality.
Documentation as Infrastructure
Documentation designed for AI agent consumption, not just human reference. Structured frontmatter on every document makes the codebase machine-parseable: an agent reads a runbook and knows what a system does, what it depends on, where its sources are, and whether it’s been audited. Maps of Content provide curated navigation. CLAUDE.md teaches an agent how to work in the repo in one read. Consistent across all three projects — the same patterns in Stars, Conductor, and Mediastack.
Multi-Agent Accuracy at Scale
When multiple AI models work across 690K lines of code, they drift without guardrails. The documentation architecture prevents this: structured frontmatter tells agents what’s authoritative, MOCs map relationships between systems, agent context files pin each model’s role and knowledge boundaries, audit status tags flag what’s been verified vs what’s stale. An agent joining the project reads one file and knows the architecture. An agent fixing a bug reads the runbook and knows the constraints. The documentation IS the coordination layer.
AI-Native Development
4 AI model families in rotation. 44 billion tokens processed. $253 total API cost over 3 months (99.7% cache hit rate). AI multiplies output; architecture and quality control are human.
Network Architecture
Multi-host WireGuard mesh with public gateway pattern. VPS protects home network. Mobile devices on mesh with ATF-gated sensor access. Pi-hole DNS filtering.
Projects
Stars Engine
Started as a feature inside Conductor on June 14, deployed the same day. Grew until it outgrew its parent repo and was extracted to a standalone project. Single ephemeris pass enables cross-tradition convergence no competitor can offer. Content-addressed archive with 152,339 charts. Concept graph with attested edges and source citations. Book of Stars personal knowledge system with progressive disclosure. Python SDK. MCP server with 13 tools.
Conductor / Cognix
5 AI model families coordinated over a cryptographically-secured message bus. LLM-based task classification with confidence gating. Cross-model review enforced by routing layer. Trust-scoped access with JIT tokens. 32 ABI handlers for uniform hardware/software interface. Self-healing infrastructure and autonomous cybersecurity response. The knowledge transferred to every project since.
Mediastack
Infrastructure, media automation, streaming, collaboration, monitoring (10 services), security (13 services including Wazuh SIEM and Falco), cybersecurity lab, OSINT stack, and automated maintenance. SSO with MFA across all services. 5-layer security architecture. restic backups with retention. Zero cloud dependency. The physical layer that Conductor orchestrates and Stars shares.
Learning Arc
First Docker Host
Ubuntu Server installed on bare metal. Learned containers from zero.
Smart Home & IoT
Home Assistant, Immich, ESPHome, CrowdSec. Learned Docker Compose, reverse proxies, DNS, and intrusion prevention.
Media Automation
*Arr stack behind gluetun VPN. Plex streaming. Traefik with TLS. Infrastructure-as-code patterns emerged.
Conductor / Cognix
Redis Streams, mTLS, HMAC, LLM classification, cross-model review, trust scoring, ABI handlers. Learned distributed systems from scratch.
Stars Engine
Started as a Conductor feature. Deployed v1 same day. Extracted to standalone repo July 10. 148 traditions, concept graph, Book of Stars, SDK, MCP server. 304K lines.
Mediastack v2
Restructured into layered architecture. Security stack, OSINT, cybersecurity lab, autonomous incident response. 77 services defined.