The brief.
A US company building a sustainable materials and clean energy business, with commercial pulp made from agricultural residues such as sugarcane bagasse and wheat straw and further lines in renewable fuels and biochemicals, wanted to find out what neural networks and generative AI could do across its operations. It needed one AI/ML engineer to design, fine-tune and deploy models, work with stakeholders to turn business needs into technical solutions, and document the work as it went.
The catch was that this was a trial. The role was part-time, fully remote and closer to a freelance engagement than a job, with a possible move to full-time later. The client had not yet decided how deeply to onboard anyone or whether the role would become permanent. It needed someone strong enough to work alone across model design, data and deployment, and comfortable with an engagement whose scope and future were still open.
- Neural networks, deep learning and ML fundamentals at mid to senior level
- 3+ years hands-on with TensorFlow, PyTorch or Keras
- Designing, training and deploying generative models (GANs, VAEs, LLMs)
- Python (R or other ML tooling a plus)
- Large datasets, preprocessing and database technologies
- Production deployment with scalability and performance in mind
- AWS, Google Cloud or Azure (advantageous)
- Clear writing and documentation for remote, part-time collaboration
- Comfort with a part-time, trial-stage engagement
The funnel.
The problem.
The arrangement was the first filter. Many strong AI/ML engineers want full-time roles with stability, and the independent ones who like part-time work often have full calendars. The pool of people who were both highly skilled and open to an uncertain, part-time engagement was a fraction of the field. We said so in the first conversation, and the funnel shows it: of 185 screened profiles, 76 remained after résumé evaluation, largely because candidates opted out once they understood the terms.
Depth was the second. AI/ML is crowded, and many profiles list frameworks and model types without having trained or deployed anything substantial. Telling an engineer who has called a hosted LLM API apart from one who has fine-tuned and shipped a model takes a technical conversation, not a keyword scan.
Independence was the third. An exploratory project comes without a tidy specification. The engineer would need to understand a business built on residue pulp and biofuels, work out where AI could help, and explain findings to non-technical stakeholders with limited overlap. Strong code was not enough if a candidate needed close direction.
What we did.
We started with a 45-minute call to pin down the role, the trial terms and what a good first few months would look like, and agreed the must-haves against the good-to-haves. That call was the largest single block of the client's time on the search.
We sourced 550+ profiles across AI/ML, deep learning and generative AI, screened to 185, and kept 76 after résumé evaluation that looked for real model building and deployment rather than listed tools. 45 reached our first internal round on neural networks, generative models, data handling and deployment. 36 cleared a second round on communication, independent working and comfort with the arrangement.
From those 36 we sent the client 13 curated profiles with our notes. The client spent about 30 minutes reviewing them and 30 minutes interviewing the candidate it chose. That candidate was hired on a contract within budget, with a path to full-time if the work grows. We handled documentation, background verification and contract setup.
What changed.
The client hired an AI/ML engineer within one week, having spent about two hours on the whole search: a 45-minute briefing call, 30 minutes reviewing profiles and a 30-minute interview. Everything else, from sourcing 550+ profiles to two internal interview rounds, sat with us.
The engineer brought hands-on TensorFlow, PyTorch and Python, had designed and built neural networks and generative models (including GANs and LLMs) for text generation and image synthesis, had handled large datasets, and held a master's degree in data science.
For a company testing what AI could do before committing to a full-time role, that meant it could start the experiment straight away, at contract cost, with an engineer able to work without supervision. If the role goes full-time later, that will be a decision about the client's needs, not a question about the engineer.
What it was worth.
What the same engagement would mean for a team that wants to test AI on its business without a full-time commitment.
From the briefing call to a signed contract.
45 minutes on the brief, 30 reviewing 13 curated profiles, 30 interviewing one candidate.
Sourcing, résumé evaluation and two internal interview rounds, all run by us.
The client tested AI on its business at contract cost, with the option to go full-time later.
How we ran it.
45-minute brief
Agreed the role, the trial terms and the must-haves, and told candidates the terms from the first conversation.
Source and screen
550+ profiles to 185, then 76 after checking for real model building and deployment.
Two internal rounds
45 tested on neural networks, generative AI, data and deployment. 36 cleared a round on communication and fit for part-time, independent work.
Client review and hire
13 curated profiles, a 30-minute review, one 30-minute interview, and a contract signed within the week.
What made this work.
Being plain about the terms early is what made a one-week search possible. A part-time, trial-stage role is easy to oversell, and a candidate who expects full-time stability will drift once reality sets in. Stating the structure in the first conversation let the wrong candidates leave early, which is why the funnel drops sharply after initial screening and why the 36 who cleared our second round actually wanted the work.
The other lesson is about the client's time. Two hours is achievable when the client only does the parts nobody else can: explaining the business, choosing between finalists who are already vetted, and making the call.