Course purpose

The Practicum is for a mid-career professional whose work is research, analysis and writing, who has no technical training, and who has reached the limit of what can be produced by hand. It teaches the applied use of two AI tools — Claude and NotebookLM — across three domains: building a durable research corpus, running recurring professional work through configured systems, and specifying and building small software tools.

It is a practicum rather than a lecture course. Every session is taught from methods the instructor uses daily in his own professional work, is conducted on the participant's own live work, and ends in something they keep. Each of the four units requires a completed piece of work before it closes, and completion is assessed against those artifacts rather than against attendance alone.

The programme is delivered as one taught 60-minute session per week across a 22-week window: three units of six sessions each, a two-session consolidation unit, and two scheduled break weeks. In addition, up to two optional support sessions of 15 to 60 minutes are made available each teaching week, for assistance with assigned practice and with the participant's continued development of the systems and tools being built. Support sessions are included in tuition, are attended at the participant's discretion, and are recorded separately from the 20 instructional contact hours against which tuition is priced.

Programme status. This is the first delivery of the Practicum. The curriculum, syllabus, assessment criteria and course materials are complete and are published in full, but no cohort has yet completed the programme. There are accordingly no participant outcomes, no completion rates and no participant testimonials, and none are claimed in this document or on the course website. The instructor's own record of applied practice, from which the course is taught, is set out below and in fuller form at www.appliedaipracticum.com/instructor.html. So that the risk of a first delivery sits with the provider rather than the participant, the first three sessions are not invoiced.

Instructor

The Applied AI Practicum is taught by Trevor Richtenburg, sole instructor. He holds a degree in mechanical engineering, which included formal programming coursework, principally Python, and has worked since April 2023 in the continuous, operational use of AI tools inside the adventure travel business he runs in Ladakh and Zanskar, in the Indian Himalaya. He is a practitioner rather than a researcher or data scientist, and has not worked as a professional software developer.

The methods taught in this course were developed to run that business, not to be taught. They include: eighteen client-facing trip sites specified, built and deployed; a server-side document-intake system for guest travel documents; an OCR receipt and reconciliation pipeline covering 457 receipts across seventeen consecutive months; generators for Indian foreigner registration and checkpoint documentation produced from passports, visas, tickets and insurance certificates; research and synthesis work supporting investment and pharmaceutical consultancy; grounded audio explainers produced with NotebookLM, one of them published; a full-day STEM teaching programme written for a village school in Zanskar; and the visual identity, published syllabus and working tool of this course. Tool families in regular use are Claude, NotebookLM, ChatGPT, Gemini and Perplexity.

The travel business in which these methods were developed, and in which they remain in daily use, holds a public customer review record: 21 reviews on Trustpilot, rated 4.5 out of 5 as of September 2026, at https://www.trustpilot.com/review/adventure-access.com. Those reviews concern travel services and not this course, and are not offered as evidence of instructional quality. They are cited only as independent, unedited confirmation that the planning, documentation and client-communication systems described above operate under live commercial conditions.

Client material from that work is not used as teaching material, reproduced, or published, including in anonymised form. A fuller account of the practice this course is taught from, together with an explicit statement of what it does not cover, is published at www.appliedaipracticum.com/instructor.html.

Learning outcomes

On completion, the participant will be able to:

1. Build and maintain a grounded research corpus that accumulates across engagements, so that reading, interviews and analysis remain retrievable after a project closes.Evidenced by: a live research notebook system and a written architecture standard.
2. Interrogate a body of sources critically: extracting structured comparisons, testing a claim against its evidence, and constructing the strongest form of a contrary position.Evidenced by: a documented prompt library for analytical work.
3. Verify AI-generated claims, and diagnose a failure by identifying which properties of the system produced it.Evidenced by: a verification checklist and failure-diagnosis sheet, applied to the participant's own output.
4. Run recurring professional work — documents, correspondence, commitments, scheduling — through configured, persistent systems.Evidenced by: configured workspaces, two working skills, and a documented workflow per recurring task.
5. Write a specification precise enough that a working tool can be built from it, and retain it as the tool's record of intent.Evidenced by: an intent file and specification, and the tool built from them.
6. Build, test, host and hand over a small single-purpose web tool, and then repeat it unassisted.Evidenced by: two working hosted tools with handover documentation.
7. Set and defend professional limits on AI use: what is never submitted to a model, what must be disclosed, and the conditions under which AI-assisted work is released.Evidenced by: a written personal AI use policy in three parts.

Session schedule

The structure below is standard. The material inside it, meaning the source sets used in Unit 1 and the tool built in Unit 3, is rebuilt around each participant's own work before the course begins.

Revision during delivery. Session content is additionally subject to revision on three grounds: the participant's stated priorities, as established before the course and in session; material change in the capability of the tools taught, which are third-party products outside the instructor's control; and the instructor's own revision of teaching material. The programme structure is not subject to revision. The number and length of sessions, the unit structure, the four required artifacts, the seven learning outcomes, the assessment method, the completion requirements and the tuition figure are fixed at enrolment and are the terms against which completion is certified. Any revision that would affect a stated outcome or a required artifact is notified to the participant in advance, and the published version is delivered instead at the participant's election.

Unit One — Research practice

Sessions 1–6 · January – early February

01	Baseline and mapping Mapping the participant's recurring work products, how each is currently made, and where the hours go. The Practicum Playbook, the course's running written record, is opened.
02	How these systems work, and how they fail Five properties that account for most observed failure: how output is generated, the boundaries of model knowledge, working memory and context limits, reliability of instruction-following, and what a citation does and does not guarantee. Then diagnosis: identifying which properties are interacting in a given failure. Placed before any productive use.
03	NotebookLM: foundations and architecture Sources, grounded responses and citation behaviour, applied to a live source set from the participant's current work; then standing against per-engagement notebooks, naming conventions, source hygiene, when to split a corpus, and disposition at the end of an engagement.
04	Capturing conversations, and what to make from them Interviews, calls and meetings from recording to transcript to source, including the consent and confidentiality questions this raises; then briefing documents, study guides, timelines and audio overviews — what each is suited to, and where presenting one would misrepresent the work beneath it.
05	Claude for analytical work Prompting for analysis rather than answers: structured extraction, cross-source comparison, testing claims against evidence, and steelmanning a contrary position. The underlying method is specification, critical evaluation and respecification — result, method and manner each stated explicitly — applied in session until one prompt is reusable.
06	The two tools together, and unit review Division of labour between a grounded corpus and an open reasoning tool, written up as a standing procedure. The unit artifact is assembled and assessed.

Unit artifact: a written research procedure, plus a completed research brief on live subject matter produced under it.

Unit Two — Professional systems

Sessions 7–12 · late February – March

07	Persistent context Projects, custom instructions and reference files: configuring context once rather than re-establishing it per conversation. Introduces skills as portable context, built in full in the following session.
08	Recurring documents, built as skills Invoices, memos, proposals and engagement summaries, encoded as skills — instruction files the tool applies automatically to matching work, owned and versioned by the participant rather than pasted per use. Each is then tested against the cases that break it.
09	Correspondence and voice Thread summarisation and drafting in the participant's own register, using a voice reference built from their prior writing.
10	Commitments, scheduling and structured data Scheduled tasks, structured checklists and recurring prompts, with explicit attention to where automation is brittle and a manual system is preferable; then converting unstructured input into structured schedule data. Also gathers the input material for the Unit 3 specification.
11	Confidentiality, judgment and disclosure What is not submitted to a third-party model when work touches client, health or government-adjacent material. Retention, provider terms and client disclosure. The resulting policy is written in three parts: responsibility in production, disclosure obligations, and the conditions under which AI-assisted output is released.
12	Unit review Every system evaluated against a week of the participant's live work, marked against pass conditions written in advance; failures repaired or deliberately abandoned and recorded as such.

Unit artifact: the configured system, documented, with the test set and pass conditions it was marked against, the results, and a written account of what was tried and what was dropped.

Unit Three — Building tools

Sessions 13–18 · April – May

13	Reading and directing code What a single-file web tool is, how to read unfamiliar code well enough to judge whether it does what it claims, and how to specify a change precisely.
14	Specification Converting a need into a written specification held as a single intent file retained with the tool: inputs, outputs, exact target format, edge cases, what the tool must refuse rather than guess, and what it was never intended to do. Requires a real example of the target format.
15	Build — first working version Specification to a running tool that handles the common case correctly.
16	Build — edge cases and validation Malformed input, ambiguous dates, multi-day and recurring events, missing fields, and correct behaviour under uncertainty.
17	Interface, hosting and handover Making the tool usable by someone other than its author — affordances, defaults, legible error states and short documentation — then static hosting, deployment, version retention, and handing the tool to a colleague so it survives without its author.
18	Second tool, participant's choice A second tool specified and built within a single session, establishing that the method transfers without instructor dependence.

Unit artifact: two working hosted tools, with the intent file and specification behind each.

Unit Four — Consolidation

Sessions 19–20 · late May – early June

19	Portfolio review All artifacts assessed against the seven outcomes; gaps identified and closed. Written assessment of unassisted capability, including what remains outside it.
20	Forward plan and handover A twelve-month maintenance and continued-learning plan; final handover of the completed Playbook.

Unit artifact: the completed Practicum Playbook and a twelve-month continuation plan.

Assessment and completion

Completion requires all three of the following:

• Attendance at 17 of the 20 main sessions. Missed sessions may be rescheduled within the delivery period; more than three unrecoverable absences results in a record of non-completion. Optional support sessions are not counted toward or against this requirement.
• Delivery of all four unit artifacts, reviewed against the stated learning outcomes.
• Completion of the capstone portfolio review in Session 19.

On completion, a Certificate of Completion is issued and signed by the instructor, recording the program title, delivery period, contact hours, units completed and artifacts produced, together with an attendance record. The certificate documents completed private professional coursework. It is not an accredited qualification and carries no academic credit.

Tuition

Item	Amount (US$)
Course tuition — The Applied AI Practicum, complete 20-session one-to-one program, inclusive of optional weekly support sessions, curriculum design, written materials, assessment and certification	3,300.00
Registration, materials and certification fees	None
Total course fee	3,300.00

Equivalent to US$165.00 per instructional contact hour, invoiced in US dollars, before any optional support sessions, which carry no charge. The first three sessions are not invoiced; the first invoice is issued after Session 3, and the schedule from that point is agreed with the participant. Where a funding scheme requires a particular schedule — including that no payment be made before a specified date — the invoice is issued to match. No deposit, registration fee, cancellation fee or instalment surcharge applies.

Software is not included in tuition. Participants hold their own subscriptions to the AI tools used in the course, purchased directly from the providers, at an expected cost of approximately US$25–60 per month during the delivery period. These costs are paid directly to the providers, are never invoiced by the instructor, and do not form part of the tuition figure above.

Documentation for reimbursement. On request and at no additional cost: an itemised invoice, this course description, an attendance record, and the completion certificate. Where a funding scheme requires that no payment be made before a specified date, the invoice schedule is issued to match.

Withdrawal and refund terms

The first three sessions are not invoiced. A participant who withdraws within that window owes nothing and there is no sum to refund. After Session 3 tuition is payable and non-refundable, as the curriculum has by then been built to the participant and a cohort place held. No deposit and no cancellation fee applies. If the instructor is unable to complete the program, the participant is refunded pro rata for every session not delivered.

The Applied AI Practicum · taught by Trevor Richtenburg · www.appliedaipracticum.com · trevor@adventure-access.com

This program is private professional training. It is not accredited, confers no academic credit, and is not affiliated with any university, government agency or AI vendor. Claude and NotebookLM are products of Anthropic and Google respectively; neither company sponsors or endorses this course. Course description issued September 2026 for the 2027 cohort, which is the first delivery of the programme.