Questions before the first build review

No on all counts. Neural Forge AI is an applied-AI studio that designs and builds custom AI systems for client organizations — not an online course, not a passive-income scheme, and not a cryptocurrency mining or GPU rental marketplace. "Forge" refers to the craft of engineering production-grade AI — not blockchain forging, metalworking as a hobby business, or gaming mod culture.

We deliver AI strategy, generative-AI applications, LLM assistants, agents, retrieval-augmented generation, workflow automation, machine learning models, data pipelines, model evaluation and MLOps as professional services. Humans remain accountable. We do not guarantee model accuracy, zero hallucinations, cost savings, revenue, ROI or that our systems will replace your staff. Outcomes depend on data quality, scope, budget and how your team adopts the tooling.

Most relationships begin with a fixed-scope discovery sprint or proof-of-concept build. You receive a written roadmap, working prototype or production deployment depending on scope. When systems are live, many clients move to a retainer for guardrail updates, evaluation harness maintenance, MLOps and incident response. Retainers are monthly with defined hours — not open-ended feature factories without change control.

We quote CAD project fees before work starts. Scope changes are discussed transparently in writing. We are an AI consultancy focused on custom delivery, not generic staff augmentation unless explicitly agreed.

Indicative ranges only — not binding quotes: discovery sprints C$20,000–C$38,000; RAG and knowledge-system builds C$90,000–C$240,000; document intelligence pipelines from C$65,000; multi-agent workflows vary with integration depth. Retainers often start around C$7,200/month for evaluation and MLOps support.

Budget depends on corpus size, security tier, number of integrations and document cleanliness. If preprocessing will dominate cost, we say so early — that honesty saves everyone time and prevents mid-project surprises.

A focused discovery loop — stakeholder interviews, data audit, architecture options, evaluation plan — typically runs two to four weeks after kickoff, assuming document access and decision-maker availability. A prototype with retrieval pipeline, basic assistant UI and benchmark on your question set often lands in seven to eleven weeks. Production deployment with full guardrails, API integration and MLOps takes longer.

Timelines are fixed in the statement of work. We do not promise instant delivery or guaranteed measurable outcomes by an arbitrary date.

We are model-agnostic pragmatists. Engagements may combine commercial LLM APIs, open-weight models, vector databases, orchestration frameworks and custom ML components. Cloud, hybrid and on-prem patterns are in scope when privacy requires them. Tool choices are documented in architecture decision records — we do not lock you into a vendor because of our preferences.

Fine-tuning, prompt design and retrieval strategy are selected based on evidence from prototypes, not trend cycles or vendor marketing.

Canadian client data is handled under PIPEDA-aligned practices: purpose limitation, consent where required, access controls, encryption in transit and at rest, retention schedules and breach notification procedures. We minimize what enters model training versus inference-only use. Cross-border processing is disclosed and contractually governed when it occurs.

Contact [email protected] for access or correction requests. See our Privacy Policy for full detail.

Custom code, prompts, retrieval configurations and integration work specified as deliverables are assigned to the client upon payment unless otherwise agreed in writing. Pre-existing studio libraries and frameworks remain ours, licensed for your use as part of the deliverable. Third-party model weights and API terms follow their providers.

We do not claim ownership of your documents or business data. Client corpora are processed only for agreed purposes and deleted or returned per contract.

Accuracy depends on source quality, coverage, chunking strategy and how questions are phrased. Retrieval-augmented generation reduces but does not eliminate hallucinations. We implement evaluation harnesses, confidence thresholds, citation requirements, refusal policies and human-in-the-loop escalation — and we report limits honestly in sprint reviews.

We do not guarantee zero errors, full autonomy or replacement of your team. AI outputs require human review for consequential decisions in finance, legal, HR, safety and customer-facing contexts.

Engagements are governed by a statement of work or master services agreement plus data processing terms where needed. Sprint reports cover progress, evaluation metrics, risks and next steps. Retainers include monthly health summaries — latency, error rates, drift signals and backlog items. Invoicing is in CAD with clear milestone triggers.

We do not use opaque "success fee" models tied to unverifiable AI outcomes. Payment terms and IP assignment are explicit before work begins.

We do not sell AI courses or certifications. We do not operate crypto mining, NFT projects or GPU rental marketplaces. We do not promise guaranteed revenue, passive income or "set and forget" automation. We do not scrape personal data without authorization. We do not train general models on your confidential data for other clients without written consent. We do not provide legal, medical or investment advice — we build software and advisory engineering services only.