50 Side Project Ideas for ML / AI Engineers in 2026
By Cristian Lascu · The Sovereign Technologist
Most ML and AI engineers have more leverage than they realise — their skills solve problems people are actively willing to pay for. The challenge is knowing where to start. These side project ideas are ranked by difficulty and revenue potential, based on what's actually working for technical professionals building income outside employment in 2026.
Starter Side Projects for ML / AI Engineers
- LLM Evaluation Framework as a Serviceadvanced3–5 months
Build a SaaS that lets teams run standardised evals on their LLM prompts and compare outputs across models. Target AI product teams who lack internal evaluation infrastructure.
high potential
- AI Readiness Audit for Mid-Market Companiesintermediate2–4 weeks
Offer a productized engagement assessing a company's data quality, infrastructure, and team capability for AI adoption. Deliver a prioritised roadmap — not a generic deck.
high potential
- Vertical AI SaaS for a Niche You Knowadvanced4–8 months
Build a narrow, specific AI tool for an industry you understand — legal contract review, construction project estimation, clinical notes summarisation. Niche beats general-purpose.
high potential
- Fine-Tuning Service for Niche Domainsadvanced4–8 weeks
Offer a productized service fine-tuning open-source LLMs on client-specific data for domain adaptation. Target regulated industries where GPT-4 privacy constraints matter.
high potential
- AI Literacy Workshop for Corporate Teamsbeginner2–4 weeks
Design and deliver a half-day or full-day workshop teaching non-technical employees how to use AI tools effectively and safely. Package it as a repeatable corporate training.
high potential
- Model Card Generatorintermediate4–8 weeks
Build a tool that takes a model's training data, evaluation metrics, and usage context and generates a standardised model card for documentation and compliance.
medium potential
- RAG Pipeline Starter Kitintermediate4–8 weeks
Build and sell a production-ready RAG (Retrieval-Augmented Generation) boilerplate with chunking, embedding, vector storage, and LLM integration pre-configured.
high potential
- AI Product Failure Post-Mortem Newsletterbeginner6–10 weeks to monetize
Write a newsletter that analyses real-world AI product failures — what went wrong, why, and what to do instead. Contrarian takes attract engineers tired of AI hype.
medium potential
- Prompt Engineering Guide by Industrybeginner2–4 weeks
Write and sell vertical-specific prompt engineering guides for common business domains (legal, finance, healthcare, customer support). Each guide includes tested prompt templates.
medium potential
- MLOps Setup as a Serviceadvanced2–6 weeks
Offer a fixed-scope engagement where you build a production ML deployment pipeline (training, versioning, serving, monitoring) for a company shipping its first model.
high potential
- AI Cost Optimisation Consultingintermediate1–3 weeks
Help companies reduce their AI API bills by optimising prompt design, implementing caching, selecting the right model per task, and batching requests intelligently.
high potential
- Synthetic Data Generation Serviceadvanced2–4 months
Generate high-quality synthetic training data for companies that lack real-world labeled examples for their specific ML task. Charge per dataset and domain.
high potential
Scalable AI Businesses
- AI Product Studio (Build + Advise)advanced3–6 months
Position as an AI product studio that both builds AI products for clients and publishes your own. The consulting work funds and validates the product work.
high potential
- AI Safety and Alignment Consultingadvanced2–8 weeks to first engagement
As regulation increases, offer consulting to companies needing AI governance frameworks, red-teaming, and bias audits. This is a genuinely underserved and growing market.
high potential
- Technical AI Content Empireintermediate6–12 months
Build a newsletter + course + consulting flywheel where technical AI content attracts an audience, converts to course sales, and surfaces high-value consulting leads.
high potential
Pro tips
- →Start with something you'd build anyway for your own use. The best side projects solve a problem you personally experience — your domain knowledge reduces risk dramatically.
- →Don't wait for a perfect idea. Ship a 'good enough' version in 4 weeks and let real users tell you what to build next. Most successful side projects look nothing like the original idea.
- →Choose a distribution strategy before you write a line of code. 'If you build it they will come' is how most side projects die. Who specifically will use this, and how will they find it?
- →Price higher than feels comfortable. ML / AI Engineers consistently undercharge. If your first 10 customers didn't push back on price at all, you're underpriced by at least 50%.
- →Protect your energy. A side project that requires 40 hours a week is just another job. Pick something that can move meaningfully on 5–10 hours a week, at least at the start.
Ready to start your sovereignty journey?
The Sovereign Technologist newsletter covers what's working for engineers building income outside employment — real strategies, real numbers, no fluff.
All topics · Read the blog · FAQ · About