Hiring an AI or machine learning developer is risky when you don't know their track record. A polished portfolio and confident pitch mean nothing if the developer can't deliver production-ready models or support your system long-term. Here's how to dig into references and reviews to separate capable developers from overpromisers.
Start with Portfolio Projects, Not Just Claims
Ask for 3–5 completed projects similar in scope and complexity to what you need. Request specifics: the problem statement, their exact role, technologies used (TensorFlow, PyTorch, cloud platforms), and measurable outcomes. A legitimate developer will have details ready and won't claim credit for infrastructure work they didn't do.
If a developer says they "built an ML pipeline" but can't explain the feature engineering, model selection rationale, or why they chose that framework, move on. Look for projects with documented accuracy metrics, inference latency, or business impact (revenue lift, cost savings, time reduction).
Reference Checks: What to Actually Ask
Don't settle for a single "yes, they're great" conversation. Contact at least two previous clients directly—ideally ones who hired the developer for work matching your needs.
Ask these specific questions:
- Did the delivered model meet the promised accuracy or performance benchmarks, and on what timeline?
- How did the developer handle edge cases or data quality issues during the project?
- Was the code production-ready, or did you need significant engineering work after handoff?
- Did they provide documentation, training, or ongoing support?
- If the project faced delays or technical blockers, how did they communicate and resolve them?
Listen for specificity in responses. A reference who can describe actual problems and solutions is credible. Vague praise is a red flag.
Review Platforms: Separate Signal from Noise
Check established platforms where AI developers list services:
- Upwork, Toptal, Gun.io: Filter by ML/AI specialization, look at project completion rates, average ratings, and—critically—recent reviews (last 3–6 months matter more than old ones).
- GitHub: Review their public repositories. Do they contribute to established ML libraries? Is their code clean, documented, and actively maintained? Inactive repos or low test coverage suggest lower standards.
- LinkedIn recommendations: Endorsements are nice but carry less weight than written recommendations that reference specific deliverables.
Watch for review padding (obvious 5-star clusters from the same timeframe) or reviews that lack technical detail. A review saying "Great communication, delivered on time" is less useful than "Built a multi-class NLP classifier with 92% F1 score; documentation was excellent."
Red Flags in References and Reviews
- Mismatched expertise: The developer has glowing reviews for web development but you need computer vision. Specialization matters.
- Vague technical outcomes: "Improved the model" without metrics. How much improvement? Measured how?
- Long gaps between projects: If there's a 2+ year silence, skills may be outdated (ML moves fast).
- Client NDA deflections: Some secrecy is normal, but a developer should still provide something specific about their approach or results.
- Single-client focus: If 80% of their portfolio is one company, they may lack breadth and adaptability.
Third-Party Verification and Testing
Before committing, run a small paid trial or proof-of-concept. Budget $2,000–$8,000 for a 2–4 week test project on a non-critical use case. This reveals whether they're collaborative, handle feedback well, and actually deliver working code.
Ask them to solve a problem similar to yours—training a classifier on your data, building an inference API, or debugging an existing model. Assess code quality, documentation, and their explanation of choices.
Pricing Context for Due Diligence
AI/ML developers typically charge $80–$200+ per hour freelance, or $150,000–$300,000+ annually for full-time roles. If someone's pricing is significantly lower, ask why. They might be junior (which is fine for certain projects) or cutting corners on testing and documentation.
Platforms like Mercoly help you compare and evaluate trusted AI and machine learning development providers in one place, streamlining the vetting process alongside your manual reference checks.
Frequently Asked Questions
Q: What should I prioritize—years of experience or recent project proof? Recent projects (last 12 months) with documented outcomes beat vague experience claims. Someone with 2 years actively shipping production ML models is more reliable than someone claiming 10 years if they can't demonstrate recent work.
Q: How do I verify that a developer's GitHub contributions are actually theirs? Ask them to walk you through a repo they claim to own—explain the architecture, key algorithms, and why they made specific trade-offs. Anyone can fork a project; understanding it requires genuine knowledge.
Q: Is a high Upwork/Toptal rating enough to hire? No. High ratings mean consistent delivery and good communication, which matters, but don't confirm technical depth for your specific challenge. Always combine platform ratings with portfolio review and reference calls.
Use these checks before signing a contract, and you'll hire developers who actually ship.