Spaces:
Configuration error
Download docs/thesis/proposal.md from AIONYX/README: direct link, hf CLI and curl.
- Browser
- Download file 3.28 kB
-
https://huggingface.co/spaces/AIONYX/README/resolve/main/docs/thesis/proposal.md
- Command line
-
hf download hf://spaces/AIONYX/README/docs/thesis/proposal.md
-
curl -L -o proposal.md https://huggingface.co/spaces/AIONYX/README/resolve/main/docs/thesis/proposal.md
AIONYX: Project Proposal (Thesis + Business)
1. Executive Summary
AIONYX is a data-driven AIOps platform designed to provide predictive reliability and cost efficiency for Kubernetes environments. It bridges the gap between infrastructure automation and machine learning, offering a system that moves beyond reactive monitoring into proactive operations.
2. The Vision: "Lean & Mean" AIOps
As a one-person project on a student budget, AIONYX focuses on a Zero-Cost, High-Impact stack. We use the best open-source tools to build a system that rivals expensive commercial alternatives (Datadog, Dynatrace).
Core Research Question (Thesis)
How can a software-defined ingestion pipeline (Dagster) and machine learning models (AIOps) be used to predict Kubernetes node failures 30 minutes before they occur?
3. Implementation Strategy (The "Solo" Plan)
To succeed as a single developer, the project is scoped around a single "Killer Feature": Predictive Anomaly Detection.
Phase 1: Local Foundation (Month 1-2)
- Environment: Use
KindorMinikube(Free) instead of expensive cloud clusters. - Ingestion: Dagster assets pulling
container_cpu_usage_seconds_totalandcontainer_memory_working_set_bytes. - Storage: Postgres (Standard) or TimescaleDB (vial Docker - Free).
Phase 2: Data & ML (Month 3-4)
- Feature Engineering: Transform raw metrics into time-series features (Rolling mean, Std dev, Rate of change).
- Model: Start with Isolation Forest (Unsupervised) to detect outliers, then move to LSTMs if time permits.
- Labels: Use your internship experience to "inject" failures (Stress tests) to create a labeled dataset for free.
Phase 3: Business MVP (Month 5-6)
- Alerting: Simple Slack/Discord integration (Free).
- Dashboard: Use Dagster's built-in UI (Dagit) as your first "product" interface—no custom frontend needed yet.
4. Zero-Cost Market Standard Stack
| Function | Tool | Commercial Equivalent | Cost |
|---|---|---|---|
| Orchestration | Dagster | Airflow/Prefect | $0 |
| Database | Postgres | Snowflake/Databricks | $0 |
| Compute | Local K8s (Kind) | EKS/GKE | $0 |
| CI/CD | GitHub Actions | Jenkins/CircleCI | $0 ($2k free mins) |
| Infrastructure | Terraform | Manual Config | $0 |
| ML Framework | Scikit-learn/PyTorch | SageMaker | $0 |
5. Business Path: From Thesis to Startup
- Open Source (The "Lead Magnet"): Release the core Dagster assets as a public repo. This builds your brand name in the DevOps community.
- Productize: Offer a "Cloud Agent" that small startups can drop into their clusters to get instant ML insights without setting up Dagster themselves.
- Target Market: Startups with < 50 nodes who cannot afford $5,000/month Datadog bills.
6. Milestones & Deliverables
- M1: Reproducible local K8s + Dagster + Postgres environment (Infrastructure).
- M2: Automated data ingestion and "Feature Store" populated with simulated failures.
- M3: Trained ML model with > 80% precision on failure prediction.
- M4: Final Thesis at Malmö University + Open Source MVP launch.
Prepared by: [User Name] Academic Advisor: Your AI Professor Date: December 2025