Claude Expert Program
Becoming properly good at Claude — sixteen live Saturdays from no experience at all to designing, evaluating and handing over Claude systems inside a business
Saturdays
Movements
Units
Live Hours
Anthropic Publishes Four Blueprints. Nobody Teaches the Road.
This is not a certification course. Anthropic's four published exam blueprints are used here as a map of what a competent Claude practitioner ought to be able to do. The course teaches to that map and then past it. If an employer later asks for a paper, the learner is ready for it. The skill is what is sold here.
"Somebody who can design an agent and cannot tell when its answer is wrong is no use to anybody."
The honest answer on the exam is given in session one. Exam registration goes through Anthropic Partner Academy and then Pearson VUE, and it requires a work email on a Claude Partner Network organisation's domain — certification is not currently open to the general public. This course prepares you fully and walks the registration flow on screen. We do not sell vouchers and we do not imply we can get you one.
No promise is made here about salary, placement or employment.
You Build Sahaya — One Client, One System
A nurse opens a pre-authorisation request and Sahaya has already read the hospital's bundle, pulled the member's policy terms, listed what is missing, and drafted a recommendation with the policy clause it rests on. Nothing is approved without a named human pressing the button, and every answer carries where it came from.
One client
A health-insurance third-party administrator that misses its 2-hour pre-authorisation SLA on about a fifth of requests
One system
Sahaya — the assistant the client's own people use, built across all four movements in one git repository
One repository you keep
It runs offline on seeded synthetic data, with an evaluation suite and a handover pack a reviewer can check
What Works at the End — and What a Reviewer Can Check
| Part of the build | What It Does |
|---|---|
| Claude Projects for three teams | Instructions, knowledge sources and an output review checklist for the non-technical teams, plus a written AI-use policy |
| The Sahaya service | Takes a pre-authorisation bundle and returns a structured recommendation with the policy clause it rests on |
| Claims-core MCP server | Deliberately read-mostly, built against spec revision 2026-07-28, with structured error responses an agent can act on |
| Coordinator and subagents | Read, check, price, synthesise and report subagents, passing context explicitly and surfacing a failure instead of swallowing it |
| Hooks carrying the hard rules | Escalation above a rupee limit, retrieved document text quarantined from instructions, and inconsistent tool output normalised |
| Policy-RAG index | A stated chunking and retrieval strategy, and a reproducible demonstration of a stale index at unchanged latency |
| Evaluation suite | At least forty cases scored for accuracy, latency, cost and safety, with an A/B record of two prompt versions |
| Team setup and handover pack | A CLAUDE.md hierarchy, path-specific rules, a pull-request-reviewing GitHub Action, and the architecture, SLA, risk and runbook documents |
Data and scope: all data in the project is synthetic and seeded in the repository. Sahaya never moves money or changes a claim's state, and the vendor claims system is never modified — it is represented by a seeded stand-in with the same API shape, so the whole thing runs offline.
Tools & Technologies
Versions verified against the official documentation on 30 October 2026.
Who It Is For
Four kinds of people, and the road is built so all four fit in one room — developers and non-developers alike, because the coding is increasingly done by the tool and the work now is building the agents around it.
Business professionals who have never written code
Operations, support, pre-sales, finance, analysts, consultants, delivery managers. The first movement assumes nothing but a browser and a paid Claude account.
Working engineers
Python or TypeScript, comfortable with REST and JSON, who have used Claude but never built on it. From the second movement on you are in a terminal and in a git repository.
Technical leads and solution architects
The people who make the design calls across Claude Code, the Agent SDK, the API and MCP, and who have to defend them. The third and fourth movements are theirs.
The people who explain it to the business
The fourth movement is deliberately not all keyboard: architecture documents, cost and SLA models, a risk register, and a trade-off explained to somebody who did not make it.
Four Movements. 35 Units. Sixteen Saturdays.
The order is judgement, then building, then design, then ownership — and that order is chosen rather than inherited. Nothing is repeated between movements; each one picks up the system the one before it left behind.
The unit of planning is the unit of work, not the session. How far a room gets in one sitting depends on what that room brings, so the trainer carries forward what is not finished and the calendar stays the sixteen Saturdays. There is deliberately no table pinning a unit to a date.
Four Blueprints. Used as a Map.
Anthropic's four published exam blueprints give the course its shape: four separate official syllabuses sequenced into one taught road. Each one is also a different job.
The map for Movement 1 - working in Claude, judging an output, and governing its use
The map for Movement 2 - the API, Claude Code, tools, MCP, agents and security
The map for Movement 3 - orchestration, enforcement, structured output and reliability
The map for Movement 4 - architecture, integration, evaluation, governance and handover
There are no prerequisites and no required courses for any of the four exams, and a learner may sit them in any order — the ladder here is a teaching order, not a gate. No domain weightings are printed on this page as a commitment: updated exam guides publish on 17 November 2026 and updated exam content takes effect on 7 December 2026 for three of the four exams, and the syllabus is re-checked against the new guides on the day they appear.
Six Things a Syllabus and a Question Bank Cannot Do
You leave holding a system, not a certificate of attendance
The Sahaya repository is yours. It runs offline, it has an evaluation suite, it has a handover pack, and it is a thing you can open in an interview or in front of your own management.
The four blueprints are a map, and the course goes past the map
Four official syllabuses sequenced into one taught road, in a teaching order rather than a weighting order. Claude Code is met early because it is the gentlest on-ramp. Evaluation is taught at architect level, where it is examined heavily.
All six published Architect Foundations scenarios are rehearsed
A sitting draws four of the six at random. Drilling all six is the difference between being prepared and being lucky, and it is not something self-paced video does.
Current to the week it runs, and it says where the official material is behind
The 200K context window, dated model IDs, the manual thinking parameter, the MCP initialize handshake, MCP sampling, HTTP+SSE as a transport and slash commands as separate files are each named and corrected.
The misconceptions are taught on purpose, by failure
The course does not describe a prompt injection; it runs one that works, then blocks it. It does not assert that hooks beat prompts; it argues past a hook and loses. It builds a subagent that invents an empty report, and then fixes it.
The honest answer on the exam is given in session one
Exam registration goes through Anthropic Partner Academy and then Pearson VUE, and it requires a work email on a Claude Partner Network organisation domain. This course sells the skill, not the badge. No voucher is sold or implied.
Learning Outcomes
By the end of the course a participant can do each of the following.
- Write a prompt that states the task, the reader and the output shape, and improve a weak answer by changing the prompt rather than retrying it
- Find a fabrication, an internal contradiction and a quietly missing exclusion in a confident answer, and name when a human must sign off
- Match a task to a Claude surface and a model tier on cost, speed and quality, and build a Project with knowledge sources and connectors
- Decide whether a use case is appropriate, apply data-sensitivity and privacy controls, and work inside an organisation AI policy
- Call the Claude API with messages, tools, streaming, vision and caching, and choose between realtime and the Batch API
- Operate Claude Code - skills, commands, agents, memory, sessions, headless mode, CLAUDE.md, settings.json - and read a trace to separate an integration fault from a model-output fault
- Author and deploy an MCP server on spec revision 2026-07-28, and choose between a built-in tool, a custom tool, a Skill and an MCP server
- Decide between a workflow and an agent for a given problem, and build the agent with the Agent SDK, a custom loop or a managed deployment
- Orchestrate a coordinator with subagents, pass context explicitly, enforce a business rule with a hook rather than a prompt, and resume and fork sessions
- Translate a business problem into an end-to-end Claude architecture, evaluate it with metrics and an A/B, govern it with a compliance register, and hand it over so another team can run it
Program Schedule
- Live sessions
- 16 sessions of 2 hours - 32.0 live hours in total
- Day
- Saturday, one every week
- Time of day
- To be announced
- First session
- Saturday 14 November 2026
- Last session
- Saturday 20 March 2027
- Breaks
- No class on 26 December 2026, 2 January 2027 or 16 January 2027
- Delivery
- Live online on Microsoft Teams, trainer-led, recorded
- Language
- English
- How a session runs
- The trainer builds and explains on screen; the room watches and asks. The first fifteen minutes go to whatever beat people during the week. Live time is explanation and demonstration, not supervised exercises.
- Between sessions
- A lab guide, the recording, and the trainer available for lab assistance - unblocking, not re-teaching.
- If a Saturday is cancelled
- There is no buffer week. A cancelled Saturday pushes the end date, and that is said at enrolment rather than discovered in February.
Start time: to be announced. The schedule runs on Indian Standard Time, and the start time is confirmed before enrolment closes. Because no clock time is fixed yet, no local-time conversions are published here; when the time is set, the equivalent in each participant's own zone — and the local day, which can differ — is published alongside it.
The Sixteen Saturdays
Simple, Transparent Pricing
One course, one price, and nothing sold separately — everybody walks the whole road. Full program access from day one, and a full refund if you are not satisfied after the first session.
Select your currency:
Claude Expert Program
one-time payment
- One course, one price — nothing is sold separately
- 16 Saturdays of 2 hours, live online and recorded
- The Sahaya repository is yours to keep
- Lab guide plus weekday lab assistance — unblocking, not re-teaching
- Certificate of completion, issued on the capstone rather than attendance
- Full refund if the first session is not for you
Joining after it has started
A late joiner pays pro-rata for the Saturdays that are left, at ₹1,000 each, and is given the recordings of every session they missed.
Full refund, no questions asked
Attend the first session of this training. If you are not satisfied, we refund your fee in full — you don't have to give us a reason.
Claude Expert Program
Frequently Asked Questions
Everything you need to know before enrolling.
The Claude Expert Program is sixteen live online Saturdays from Rathinam Trainers that take a person from no experience at all to designing, evaluating and handing over Claude systems inside a business. It is taught through one project — Sahaya, an assistant for a health-insurance third-party administrator — built across four movements in one git repository: judgement, building, design and ownership.
No. This is not a certification course and no badge is promised. Anthropic's four published exam blueprints are used as a map of what a competent Claude practitioner ought to be able to do, and the course teaches to that map and then past it — so a learner who is later asked to certify is ready. Certification itself is Anthropic's, delivered by Pearson VUE and badged through Credly. Registration starts on Anthropic Partner Academy and requires a work email on a Claude Partner Network organisation's domain, so it is not currently open to the general public. No exam voucher is sold or implied, and no guarantee is made that a learner can register.
Both developers and non-developers. Four kinds of people fit in one room: business professionals who have never written code (operations, support, pre-sales, finance, analysts, consultants, delivery managers), working engineers, technical leads and solution architects, and the people who have to explain all of it to the business. The first movement assumes nothing but a browser and a paid Claude account.
Sixteen sessions of two hours — 32.0 live hours in total — one every Saturday from Saturday 14 November 2026 to Saturday 20 March 2027, with no class on 26 December 2026, 2 January 2027 or 16 January 2027. The schedule runs on Indian Standard Time and the start time of day is to be announced; it is confirmed before enrolment closes. Because no clock time is fixed yet, no local-time conversions are published.
Fully live online, on Microsoft Teams, trainer-led and in English. Every session is recorded and the recordings stay available to the cohort.
The course fee is INR 13,500 including GST (or the equivalent in USD, EUR, GBP, CAD, AUD, SGD or AED depending on your region). It is sold as one course and nothing is sold separately — there are no levels to choose between and no modules to buy. Somebody joining after the course has started pays pro-rata for the Saturdays that are left, at INR 1,000 each, and is given the recordings of every session they missed. Payments are processed via Razorpay.
Yes — a full refund, no questions asked. Attend the first session of this training, and if you are not satisfied, we refund your entire fee. You do not have to give a reason. Just email rathinam@rathinamtrainers.com or message us on WhatsApp, and the refund goes back to your original payment method through Razorpay, usually within 5-7 business days.
Sahaya — the assistant the client's own people use. By the end a reviewer can open the repository and check: a documented Claude Projects setup for three teams with a written AI-use policy; a running service that returns a structured pre-authorisation recommendation with the policy clause it rests on; a claims-core MCP server on spec revision 2026-07-28; a coordinator agent with read, check, price, synthesise and report subagents; hooks that make the client's rules real; a policy-RAG pipeline with a stated chunking and retrieval strategy; an evaluation suite of at least forty cases with an A/B record; Claude Code configured for a team with a pull-request-reviewing GitHub Action; a handover set with an architecture document, SLA and cost model, risk and compliance register and runbook; and a per-learner certification plan.
For the first movement, nothing beyond a modern browser and a paid Claude account on claude.ai (Pro or above). From the second movement on: a terminal and comfort using one, Python 3.13 with uv, Claude Code installed with the native installer, Git, an Anthropic Console API key, a throwaway git repository, Node 22 LTS only if you want to follow the one TypeScript MCP example, and a small but real API budget. Knowledge-wise: for Module 1 none; from Module 2 working Python, basic git and the ability to read REST and JSON documentation. No prior Claude API, Agent SDK, MCP or agent-building experience is assumed.
Claude Haiku 5.5, Sonnet 5.5 and Opus 5.5, with a 1M-token context window and adaptive thinking steered by an effort parameter; the Claude API — Messages, tool use, streaming, vision, thinking, prompt caching and the Batch API; the Claude Agent SDK for Python with built-in tools, hooks, subagents, permissions, sessions, skills, commands and memory; Claude Code including the CLAUDE.md hierarchy, settings.json, skills with context: fork, path-specific rules, plan mode, headless and streaming modes and GitHub Actions; the Model Context Protocol on spec revision 2026-07-28; and the service and data layer — Python 3.13, uv, FastAPI with Pydantic v2, SQLite with sqlite-vec, pytest, Git and GitHub Actions. Versions were verified against the official documentation on 30 October 2026.
The trainer builds and explains on screen; the room watches and asks. The first fifteen minutes of every session go to whatever beat people during the week. The practice happens in the week, with a lab guide and the recording, and the trainer available for lab assistance — unblocking, not re-teaching. Live time therefore carries explanation and demonstration, not supervised exercises.
Yes — a Rathinam Trainers and Consultants certificate of completion for the Claude Expert Program, issued on the capstone rather than on attendance. The capstone is the Sahaya repository and its handover documents, assessed against the ten end-state items in the project section of this page. This is the trainer's own assessment and it is not an Anthropic credential.
Exam vouchers, exam fees and sitting the exam. Writing to the client's core claims system — Sahaya never moves money or changes a claim's state. Production hosting, scaling and high availability — everything runs on a laptop and in GitHub Actions, and capacity and SLA numbers are modelled on paper. Real personal data — all project data is synthetic and seeded. Fine-tuning and model training. Vector-database operations at scale. Front-end development. Languages other than Python for the build. Claude Cowork as a taught surface. And no domain weightings are printed as a promise.
No. Rathinam Trainers and Consultants Private Limited (CIN: U80900TZ2013PTC019500) is an independent technology training and consulting company. We are not affiliated with Rathinam Group of Institutions, Rathinam College of Arts and Science, Rathinam Technical Campus, Rathinam Institute of Management, or any other entity in the Rathinam Group — these are entirely separate legal entities. Rathinam Trainers is located at Irugur, Coimbatore 641103; the Rathinam Group of Institutions is located at Eachanari, Coimbatore 641021, a different part of Coimbatore.
Ready to Get Properly Good at Claude?
Sixteen Saturdays, 14 November 2026 to 20 March 2027, live online and recorded. One course, one price, and a repository you walk out holding.
Rathinam Trainers & Consultants Private Limited
Session start time is to be announced and is confirmed before enrolment closes. This course sells the skill, not the Anthropic badge: exam registration requires a work email on a Claude Partner Network organisation's domain, and no voucher is sold or implied.