HelixTrack beta
GoGinHTTP/3 QUICPostgreSQL+8
A free-world JIRA alternative; the flagship of the Helix-Track line.
// AI engineer
Miloš Vasić
AI engineer building verifiable AI-development systems: multi-provider LLM infrastructure, autonomous agents and orchestration, and the governance and QA layers that keep them honest.
Go
backbone language
140+
repositories, one constitution
33
products, evidence-backed
// summary
I design and build the software that turns LLMs into dependable products.
I work primarily in Go, with Kotlin / Kotlin Multiplatform, TypeScript/React, Python, Swift, and Shell as the job demands. My output is organized as a fleet rather than a pile of one-off apps: large product applications sitting on top of dozens of small, decoupled, independently-tested modules, all inheriting a shared engineering Constitution as a Git submodule.
That means fixes propagate across everything, new products assemble from proven parts, and every advertised capability is backed by an evidence-producing test. My north star is a single rule — a feature is finished only when a real user can use it and there is captured evidence to prove it. This page moves from the overview to the individual projects; each links to its full product page.
// my work across the Helix family
The Helix line, in priority order.
The Helix line spans the AI-development lifecycle. Each card links to its full product page.
HelixAgent beta
GoGinLLMsVerifierPrometheus+6
An ensemble LLM service with multi-round model debate and verification-based provider selection.
HelixCode beta
GoRedisSSHMCP+2
A distributed AI development platform that divides work across SSH-managed workers with automatic checkpoint/rollback.
HelixLLM beta
GoHTTP/3 QUICllama.cpp+5
A single binary with six modes serving OpenAI- and Anthropic-compatible APIs over HTTP/3, with local llama.cpp inference and a scored fallback chain.
HelixCluster in-development
GoZiggRPCRaft + SWIM+7
A distributed operating system for AI compute, from datacenter GPUs to edge handhelds.
LLMProvider beta
Gonet/httplogrus+2
One interface over 43 providers with circuit breakers, retries, and health baked in.
LLMOrchestrator beta
GoJSON-lines / stdioCircuit breaker+2
One control plane for every headless CLI coding agent.
LLMsVerifier beta
GoGinSQLite + SQLCipher+6
Verify, monitor, optimize: the single source of truth for LLM/provider/verification metadata.
// my work across vasic-digital utils
Product-grade tools I built and maintain.
Catalogizer production
GoGinReact+10
Multi-protocol (SMB/FTP/NFS/WebDAV/local), encrypted, self-hostable media collection management with a Go/Gin API and React UI.
Courses-Creator production
GoReactElectron+8
A markdown-to-video course pipeline with multi-LLM enrichment, TTS, and desktop/mobile/web players.
VisionEngine active
GoGoCV / OpenCVLLM vision
A decoupled Go toolkit fusing classic computer vision with multi-provider LLM vision for UI analysis and navigation graphs.
DocProcessor active
GoLLM agentsHeuristic parser
Turns documentation into a verifiable feature map for QA automation (LLM or heuristic extraction).
Docs Chain active
GoDAGSQLitefsnotify
A content-hashed, bidirectional, atomic document/DB sync engine.
Herald active
GoShellLLM intent
Reliable multi-channel notifications with natural-language, three-tier intent resolution.
task_bridge scaffold
GoSQLitewebhooks
A decoupled, bidirectional task/board sync engine (P1 scaffold; sync logic in progress).
Reusable Module Suite scaffold
Go (digital.vasic.*)KMP+1
The digital.vasic.* "standard library" of infrastructure, AI-primitive, and guardrail modules.
// infrastructure heritage (Server Factory)
Predating the AI line — my DevOps toolchain.
Mail Server Factory production
KotlinShellDocker+4
Declarative JSON → fully-provisioned Dockerized mail servers, reporting 439 passing tests and a clean SonarQube gate.
Server Factory Core stable
KotlinShellGradle
The Server Factory Core Framework every service factory builds on.
Qemu-Utils active
ShellQEMUISO images
QEMU VM-image tooling — download, run, network, and publish machine images as artifacts.
Parallels-Utils active
ShellPython 3Parallels
Compress, publish, and reuse Parallels VM images across every machine.
// how I work — governance and QA first
The discipline that underpins the products.
Before the products, the discipline: I don't ship green checkmarks — I ship AI systems with the evidence that they actually work.
HelixConstitution shipped
A universal, inheritable engineering rulebook, distributed as a Git submodule across a 140+-repository fleet. It codifies anti-bluff evidence gates, false-positive immunity, data/host safety, and coverage rules; projects may tighten but never weaken it, and each governance gate is paired with a mutation test that proves the gate itself works.
HelixQA beta
Anti-bluff QA orchestration that runs written test banks and fully-autonomous, LLM-and-vision-driven QA sessions across Android, Android TV, Web, and Desktop, scoring a PASS only when it has captured runtime evidence.
// technologies
The stack I actually build on.
Languages
GoKotlin / KMPTypeScriptPythonSwiftShell
AI / LLM
43+ providersMCPRAGVector DBsAgent orchestration
Backend
GingRPC + ProtobufHTTP/3 (QUIC)WebSockets
Data & Infra
PostgreSQLSQLite / SQLCipherRedisDocker / K8sPrometheus
// contact
In one line.
I don't ship green checkmarks — I ship AI systems with the evidence that they actually work, and the governance that keeps them that way. Open to senior AI/platform engineering roles worldwide.