1. AI Discovery & Strategy
Before anyone touches a GPU, we frame the problem in terms your leadership and engineers can agree on. Our discovery sprint inventories data sources, interviews domain experts, maps existing automation and separates trainable signal from wishful thinking. You receive an AI strategy document with a prioritized roadmap: which use cases deserve predictive ML, where retrieval-augmented generation beats fine-tuning, and where the honest answer is better labelling rather than a bigger model. We define measurable outcomes, document data gaps and quote CAD project fees in plain language — never guaranteed ROI, always explicit about what the crucible can hold. Typical engagement runs one to three weeks.
Indicative range: C$9,000–C$20,000
2. ML & Predictive Models
Churn, demand forecasting, fraud scoring, equipment failure, routing optimization — when your history supports prediction, we train machine learning models that respect seasonality, class imbalance and regime change. We are direct when data is too sparse or too dirty to forecast reliably; sometimes the right answer is better monitoring, not a neural net. Where models earn their place, we deliver feature pipelines, training code, evaluation harnesses and documented accuracy limits stated before launch. Responsible AI review covers bias checks on protected attributes where applicable. Production deployment includes API integration and runbooks your team can maintain.
Indicative range: C$50,000–C$130,000
3. Generative AI & LLM Systems (RAG)
General-purpose chatbots rarely survive first contact with your policy PDFs. We design retrieval-augmented generation pipelines with chunking strategy, embedding choices, metadata filters and access control aligned to your org chart. AI assistants cite sources, flag low-confidence retrievals and route edge cases to humans rather than guessing. Prompt engineering is documented and versioned; guardrails block outputs without grounding. We integrate with Slack, Teams, CRM and internal portals via API. PIPEDA-compliant data handling for Canadian client corpora is standard — your documents stay in environments you control. Proof of concept from C$40,000; full production RAG from C$85,000 upward depending on corpus size and compliance requirements.
Indicative range: C$40,000–C$175,000
4. Fine-tuning & Domain Adaptation
Fine-tuning is not a default — it is a decision we justify with retrieval baselines and cost analysis. When domain vocabulary, tone or task format genuinely requires weight updates, we prepare curated datasets, run supervised fine-tuning with documented splits and ship regression harnesses that catch drift before users do. LoRA and full fine-tune paths depend on model licence terms and your inference budget. We temper checkpoints under adversarial inputs and latency constraints, not just validation accuracy. Shortcut adaptation is how models crack when production traffic arrives. Engagements include dataset review, training pipeline, evaluation and handover documentation.
Indicative range: C$35,000–C$95,000
5. MLOps & Production Deployment
Shipping is when the real heat starts. We implement MLOps pipelines — versioned models, automated evaluation, drift detection, rollback paths and observability on inference latency and cost. CI/CD for ML includes data validation gates so upstream schema changes do not silently poison predictions. We deploy to AWS, Azure, GCP or on-prem according to your security posture; tool choice follows your requirements, not our preferred vendor badge. Documentation covers incident response, retraining triggers and who owns what after handover. Retainer support keeps pipelines current as dependencies and data shift.
Indicative range: C$45,000–C$140,000 · Retainers from C$7,000/month
6. Evaluation & Human-in-the-loop
Models without evaluation are decorative. We build golden-set harnesses, LLM-as-judge workflows where appropriate (always with human audit samples), and business-readable dashboards for precision, recall and task-specific metrics your leadership can interpret. Human-in-the-loop design defines when operators must approve, override or escalate — not as a footer disclaimer but as architecture. We run red-team sessions on generative outputs, document failure modes and train client teams on override procedures. Audit-only evaluation engagements suit teams with existing models that never survived proper tempering. Ongoing evaluation retainers catch regression before customers do.
Indicative range: C$15,000–C$55,000 audit · C$8,000–C$22,000/month retainer
