Markus Isaksson

Junior AI & full-stack developer in Stockholm.
LLM agents with guardrails, RAG and agent evaluation, and CI regression gates, in TypeScript and Python.

AI Engineering

Featured AI projects

Small, tested, offline-runnable projects about making LLM systems safe and measurable. 16 freeCodeCamp certifications (3,600+ hours) and 20+ published Agensi agent skills with 380+ installs.

agent-action-gate

A strict-TypeScript gate between an LLM agent and a backend: schema, role and limit checks, idempotent retries, human approval above a threshold, a hash-chained audit log, and a fullstack web-operations demo. Tested in CI.

Open showcase -> · Repository ->

rag-vector-eval

Offline RAG evaluation: Chroma retrieval, faithfulness, recall and MRR metrics, SQL data-quality checks on load, and a regression gate that runs in CI.

Repository ->

fastapi-ai-automation-gateway

FastAPI gateway for LLM automation jobs: hashed API keys, Swedish PII redaction (email, phone, personnummer) before any model call, append-only audit log. Tested in CI.

Repository ->

agent-trace-evals

Stochastic, trace-based evaluation of tool-using LLM agents, plus a RAG module (BM25, dense, hybrid, rerank, refusal gate) with honest end-to-end results. Tested in CI.

Open showcase -> · Repository ->

llm-json-fixer

MCP server that deterministically repairs malformed LLM JSON, with a held-out evaluation set.

Repository ->
Showcases

More agent projects

Short, self-contained demos with live sample output and a linked repository.

hybrid-audit-ts

Deterministic rules first, LLM findings second: provider output is validated and merged with rule findings into one deduplicated action list, in strict TypeScript.

Open showcase ->

inbox-agent-ts

A background agent that turns email/CRM events into typed knowledge-graph memory and scheduled follow-ups, in strict TypeScript.

Open showcase ->

agent-skills-enablement

A tool-agnostic kit for rolling out AI assistants, Agent Skills and MCP connectors: skill portability lint, adoption KPIs, governance intake, workshops, and a strict-TypeScript skill/tool router with a reference architecture.

Open showcase ->
Infra & Full Stack

The Full Stack Observatory

A 12-layer production reference architecture built to be reviewed, explained, and run locally: Express API, Postgres, Nginx, Terraform, Prometheus and Grafana. Implementation repo · Recruiter summary · 5-minute walkthrough

For Recruiters & Hiring Managers

How to evaluate the Observatory quickly

This portfolio is meant to show how I think about systems, not just isolated frameworks. If you only have a few minutes, use the path below instead of browsing the repository blind.

Architecture Security Observability Deployment Documentation

1. Read the short summary

See what the project is, what was implemented, and what technical signals it is intended to demonstrate.

Open recruiter summary ->

2. Open the public explainer

The interactive site explains the same system visually and makes the 12-layer model easy to inspect.

Open the explorer ->

3. Jump to the real code

Use the implementation map below to inspect the API, database, server configs, security layer, and observability wiring.

Open implementation map ->

4. Run the stack locally

The repo includes a one-command local demo with API, Postgres, Nginx, Prometheus, Grafana, and alerting.

Open the guided demo ->

The 12 Layers

Layer 1 - Frontend

Credentials: 16 certifications, 3,600+ hours

Layer 2 - Backend / APIs

Express + Zod + shared metrics

Layer 3 - Database

PostgreSQL + migrations

Layer 4 - Servers

Nginx + PM2 + systemd

Layer 5 - Networking

TCP handshake, teardown, and full lifecycle visualizer

Layer 6 - Cloud Infrastructure

Three architecture scales with Terraform IaC

Layer 7 - CI/CD Pipelines

This site deploys itself and visualizes the pipeline

Layer 8 - Security

Helmet, CSP, rate limiting, and Zod

Layer 9 - Containers

Multi-stage Dockerfiles and runtime packaging

Layer 10 - CDN

Cache hit/miss simulator and Cache-Control builder

Layer 11 - Monitoring

Prometheus + Grafana live

Layer 12 - Backups & Recovery

DR planner with RPO/RTO calculator and runbook

Implementation Map

Where the real system lives

If you want code instead of narrative, these are the highest-signal entry points inside the implementation repository.

Layer 2 - API

Express app assembly, middleware order, route composition, validation, and shared error normalization.

View apps/api-server ->

Layer 3 - Database

Schema constraints, migrations, and seed data instead of hiding data design behind an ORM only.

View migrations ->

Layer 4 - Servers

Nginx, PM2, and systemd configs for the non-container deployment path.

View server configs ->

Layer 8 - Security

Focused middleware examples and tests for headers, CORS, rate limiting, and XSS-related concerns.

View security layer ->

Layer 11 - Monitoring

Metrics package, Prometheus wiring, Grafana dashboards, and health-check patterns.

View metrics package ->

System Boundary

The full local runtime is visible in one place: compose orchestration, container wiring, and provisioning config.

View docker-compose.fullstack.yml ->

See it running live

One command spins up the entire stack with real metrics and observable runtime behavior.

Terminal startup

One-command startup

Grafana dashboard

Live Grafana dashboard

Prometheus targets

Prometheus targets

API health page

Healthy API + Nginx

Try the full stack on your machine in seconds

git clone https://github.com/MarkusIsaksson1982/fullstack-observatory.git && cd fullstack-observatory && npm install && npm run compose