Matin Alijani

I build systems that tell you when they're failing.

I'm Matin Alijani, a backend and AI engineer. I take products from first commit to production: the machine learning, the API, the payments and the servers. Then I add the monitoring that catches problems before your customers do.

Available now for remote roles or relocation. Based in Tbilisi, GMT+4.

Most engineers hand you one piece. I hand you the whole system.

The model, the API, the payment layer, the servers and the alert that fires at 3 a.m. For nearly three years I was the only engineer on a platform, so I know what really breaks in production, and how to make sure it breaks only once.

6+ years of PythonDjango, FastAPI, Celery, PostgreSQL
500+ Linux serversProvisioned automatically, 99.9% uptime
100,000+ usersServed through a network I built and ran
100/100 LighthouseOn a site that drew 200K+ views in 30 days

Work I designed, shipped and kept running.

A production platform, and two open-source tools built around one idea: automation is only as safe as what it refuses to do.

ABEBIT

An AI market-intelligence platform, built and run alone.

Rules decide. The model only scores.XGBoost and SGD rate signal quality but can never override the rule layer, so every decision can be explained afterwards.

Built to be checkedA validation gate rolls back any model that gets worse, and every outcome goes into a public append-only ledger (2,000+ so far).

Built for volume103 assets across three timeframes, fed by 50+ concurrent data sources at sub-second latency. Python and Rust on Django, Celery and PostgreSQL.

Payments without custodySubscriptions and Bitcoin payments run through a self-hosted wallet. The platform never holds user funds.

Visit abebit.com
Open source

IGuard

Incident response for iGaming infrastructure.

  • 40 alerts become one incident. Deduplication is enforced by a database unique index, not just application code.
  • Safe recovery. A YAML allowlist, dry run first, cooldowns, a circuit breaker and idempotency keys. Databases are never auto-restarted.
  • Missing data is never zero. If Prometheus is down, the dashboard says "no data" and no incident opens or closes.
  • Secure by default. Argon2id, rotating refresh tokens with reuse detection, CSRF protection, RBAC and an audit log.

FastAPI, SQLAlchemy 2, PostgreSQL, React and TypeScript, Prometheus, Loki, Grafana. 341 backend, 42 frontend and 8 browser tests.

View on GitHub
Open source

KubiDoor

Kubernetes monitoring with safe, limited recovery.

  • Finds and explains. Detects CrashLoopBackOff, OOMKilled, stuck pods and degraded rollouts, groups them into incidents and reports the probable cause with its confidence.
  • Two actions only. Restart a deployment or delete one pod. No nodes, volumes, Secrets, RBAC or shell.
  • Off by default. Dry run, human approval, namespace allowlists, cooldowns and rate limits. It re-checks the exact object and verifies the fix afterwards.
  • Honest under uncertainty. If the Kubernetes API is unreachable, it never marks incidents as resolved.

Python, FastAPI, SQLAlchemy, Docker, Prometheus, Grafana. 236 tests, strict typing and linting in CI.

View on GitHub

Verified: test suites and the Prometheus, Alertmanager and Loki configs against the official validators. Not yet: a full docker compose run of the whole IGuard stack, documented openly in the repo.

Three rules I learned by running production alone.

Cap the model.

A model can rank a candidate, but it should never override the rules beneath it. A system you can't explain afterwards isn't production-ready, however accurate it is.

Listen for silence.

AI systems fail quietly. After a pipeline stopped without raising a single error, I built a watchdog that alerts when output disappears, not only when exceptions appear.

Prefer the boring answer.

If a rule can solve the problem, use the rule. It is cheaper, faster and provably correct. Reach for a model only when a rule can't do the job.

Where I've done it, and with what.

Founder and backend engineer, ABEBIT

Sole engineer across ML, backend, payments, infrastructure and on-call, in Python and Rust.

Full-stack developer (contract), Zara Vet

Delivered a complete clinic platform in under three months.

Co-founder and software engineer, BOOF ABI PARSEH

Shipped the MVP, cut API response times by about 45% and reduced technical debt by about 30%.

Founder and software engineer, CastVPN

Built the encrypted network and the fleet behind it.

Backend developer (contract), Laptori

Delivered 100+ projects at a 95% satisfaction rate.

BackendPython, Rust, Django, FastAPI, Celery, PostgreSQL, Redis, REST APIs
AI and dataXGBoost, scikit-learn, LangChain, LangGraph, Claude and OpenAI APIs, backtesting
InfrastructureDocker, Linux, Nginx, CI/CD, Cloudflare, Kubernetes fundamentals
SecurityOAuth 2.0, JWT, RBAC, OWASP Top 10, server hardening
Front endJavaScript, HTML, CSS, Flutter

Have a system to build, or one that keeps failing quietly?

I'm open to backend, AI systems and platform engineering roles, remote or on-site, and happy to relocate. Tell me what you're building.