It goes alongside the proposal you accepted, and it is meant to keep the project shippable: clear inputs, clear acceptance checks, and a handover another engineer can run without guessing. If something is unclear or reality changes, I will write it up and get your approval before I change course.
What I will deliver
Machine learning build covers the practical work needed to turn your current process and data into a model you can run again next month, not a one-off demo. I start with a data audit and leakage check, then lock down labeling rules, build a baseline and error analysis, agree offline evaluation thresholds, and finish with a deployment hook plus monitoring.
Your deliverables are listed below. Anything not listed is not included unless you and I add it as a written change, including new prediction targets, new data sources, or different acceptance thresholds.
- Feasibility and data readout
- Labeling spec and gold set
- Feature store and training dataset
- Training and evaluation package
- LLM/RAG prototype and retrieval eval
- Deployment and monitoring setup
Payment
For setup work, I invoice 30% to start. I begin once that invoice is paid and I have enough access or exports to make progress. I invoice the remaining 70% when I hand over the build, which includes the code and run instructions, the training and evaluation package, and the agreed documentation.
Ongoing work runs on a retainer. I invoice at the start of each month for that month’s work. Every invoice is payable within 7 days of its date. If an invoice goes past due, I pause work and scheduled delivery dates move with the pause. I pick back up once the account is brought current.
Revisions and changes
A revision round is one cycle where you review what I delivered, you send me a single consolidated set of notes, and I apply them. For machine learning build, you get two revision rounds on the deliverables, focused on things like labeling edge cases, evaluation cutoffs, and monitoring metrics.
If you want changes beyond those rounds, or if the plan changes after I start, I will price the change and explain what it does to timeline and risk before I do the work. Common scope triggers are a new target definition, adding a new vendor feed, or changing the acceptance thresholds after we agreed them.
Who owns what
Once the final invoice is paid, you own what I created specifically for you in this engagement. That includes the code I wrote for the build, the trained model artifacts I produced, the training and evaluation package with reproducible runs, the labeling spec, the gold set, and dataset version notes so an in-house hire can retrain and audit what changed.
What I keep are my pre-existing tools, templates, scripts, and general know-how, even if I use them to produce your deliverables. If I include third-party libraries or services, they stay under their own licenses and terms. I will call those out in the handover notes where it matters for running or retraining.
Portfolio use
I may show the finished work as part of my portfolio so future clients can understand the kind of problems I solve. In this trade, that usually means a short case study about the objective, the evaluation approach, and what the system looks like in production, not your raw data.
If you do not want me to share anything about this project, tell me in writing and I will keep it private. If you are okay with a portfolio piece but need constraints, I can work within them, for example: anonymizing your company name, avoiding screenshots of internal tools, or waiting until after your feature ships.
Retainer rhythm
This engagement is bought as a retainer. Each month, I plan work in a monthly block: monitoring review, drift checks, performance against the thresholds we agreed, and retraining or fixes when the numbers cross the line or the data distribution shifts. I invoice at the start of the month, and I use that month’s time for the work we agree for that period.
The retainer renews month to month unless either of us ends it. I will not assume a notice period. If you want one, or if you need me to reserve a specific cadence, you and I will agree the notice period in writing so expectations match reality.
Limits and assumptions
I am responsible for doing the work with care, keeping the project reproducible, and documenting what I did and how to run it. I am not responsible for decisions you make using the model, or for business outcomes driven by factors outside the data and system I built.
Models change in performance when data changes. Monitoring can show what changed, but it cannot prevent change on its own. I also cannot promise that a model will meet a particular accuracy level unless we have agreed offline evaluation thresholds and your data in production matches what we evaluated. If your inputs, policies, or upstream systems change, I may need to re-check leakage, recalibrate, or retrain.
Ending the work
Either of us can end the engagement by telling the other in writing. When that happens, I will stop taking on new tasks after the end date we agree, and I will invoice for work completed up to that day, including any committed time in the current monthly block.
After final payment, I will hand over what is complete at that point: the repo or code package, run instructions, current model artifacts, evaluation results, and any notes needed to continue. If access is being removed, give me a short window to export the agreed handover materials from your tools. I will also return or delete your data from my systems where I have any copies, as practical for the way you provided it.







