[ pane 00 / solutions ]
Three programmes, stated workloads, published syllabi
Each course is scoped to a specific part of AI development work. Prerequisites are listed as self-check exercises so you can decide for yourself before committing to anything.
< Back to HomeEach Synapsia programme is built around code that runs, sessions that are live and feedback that is written. The sequence within each course follows the same logic: you read the component before you run it, and you run it before a library wraps it for you.
The prerequisite system is designed to be honestly used. A self-check exercise is published alongside the syllabus for every programme. If you work through it and find two or three places where you get stuck, that is useful information — not a reason to push through, but a reason to look at whether the time is right.
The Practitioner Programme adds a pre-enrolment technical conversation. Either side can decline to proceed. This is not a test — it is a structured way to decide together whether the programme fits the person and the person fits the programme.
- Code in repositories compiles and produces correct results under the stated Python and library versions
- Feedback on exercises is written within two working days of submission
- Sessions are recorded and available within 24 hours for cohort members
- Syllabi state the scope: excluded items alongside every topic list
- Stated workloads are updated after each cohort if actual hours consistently differ
→ next pane: [ 02 ] foundations course
Foundations: How a Transformer Actually Works
A six-week course that builds a small transformer from an empty file, one component per week. By the final session the cohort has a model that writes recognisable (if bad) text and a mental model that makes the literature readable.
// scope: included
- Tokenisation and vocabulary — why it shapes everything downstream
- Embeddings and positional encoding
- Attention written out by hand before any library call
- Multi-head attention and the intuition for why it works
- Layer normalisation and residual connections
- Training loop, loss, backpropagation
- Sampling strategies and temperature
// scope: excluded
Fine-tuning large pre-trained models, RLHF, retrieval-augmented generation, inference infrastructure. Those topics are covered in later programmes.
// self-check: can you honestly say yes to these?
- Comfortable writing Python: functions, loops, file I/O, classes
- Willing to sit with matrix shapes until they make sense
- Can run a Jupyter notebook and read an error traceback
- Some exposure to NumPy (or willingness to pick it up quickly)
// workload per week
// what is included
- Live sessions (2 per week × 90 min)
- Session recordings
- Weekly exercise sets with written feedback
- Step-by-step repository (yours to keep)
- Cohort channel
→ next pane: [ 03 ] deployment course
// self-check: can you honestly say yes to these?
- Comfortable with Python and the command line
- Basic Docker familiarity — can build and run an image
- Have run ML model inference at least once, even locally
- Some exposure to HTTP APIs (requests, status codes)
// workload per week
// what is included
- Live sessions (1 per week × 2 h)
- Session recordings
- Reference deployment repository
- Load-test script
- Written feedback on your deployment
Short Course: Deploying a Model Behind an API
A three-week practical course on the gap between a working notebook and a service other people depend on. You deploy one small model to a service that survives a load test.
// scope: included
- Packaging and containerisation with Docker
- Request batching and warm starts
- Latency budgets and where the time actually goes
- Load testing and graceful failure
- Logging what you will wish you had logged
- Cost per request and monitoring for quality drift
// scope: excluded
Kubernetes, multi-region deployment, model training pipelines, GPU cluster management. The course is scoped to the inference side of a single-region deployment.
→ next pane: [ 04 ] practitioner programme
A seven-month, part-time programme for developers who want the whole arc rather than a single topic. Roughly seventy per cent of the hours are project work. Five artefacts are reviewed line by line in writing by practising engineers, with fortnightly one-to-one office hours.
The programme closes with a capstone the learner defines and defends: a system that must run end to end for a month and be handed to a reviewer who did not build it.
// scope: excluded
Academic research methods, paper writing, hardware design, regulated professional activities. The programme is technical education, not a research degree or professional qualification.
// topics across 28 weeks
Weeks 1–6: Numerical foundations and models from scratch›
Weeks 7–12: Transformer end to end›
Weeks 13–16: Adaptation of small open-weight models›
Weeks 17–20: Retrieval systems and tool-using agents›
Weeks 21–24: Serving, cost and monitoring›
Weeks 25–28: Capstone›
// workload per week
→ next pane: [ 05 ] comparison
Which programme fits where you are now
| Feature | Foundations | Deployment | Practitioner |
|---|---|---|---|
| Duration | 6 weeks | 3 weeks | 28 weeks |
| Weekly hours | ~8h | ~7h | 12–15h |
| Live sessions | 2/week × 90 min | 1/week × 2h | 2/week |
| Written feedback | Weekly exercise sets | On deployment | 5 artefacts, line by line |
| Office hours | — | — | Fortnightly 1:1 |
| Capstone | — | — | ✓ defined and defended |
| Compute credit | — | — | ✓ included |
| Pre-enrolment conversation | Optional | Optional | Required |
| Best for | Developers new to the mechanism | Developers who need to serve a model | Developers who want the full arc |
| Fee (RM) | 1,290 | 870 | 4,650 |
→ next pane: [ 06 ] fees
Fees
6 weeks
RM 1,290
one payment at enrolment
- ✓ Live sessions (12 sessions)
- ✓ Session recordings
- ✓ Weekly exercise sets + written feedback
- ✓ Step-by-step repository
- ✓ Cohort channel
3 weeks
RM 870
one payment at enrolment
- ✓ Live sessions (3 sessions × 2h)
- ✓ Session recordings
- ✓ Reference deployment repository
- ✓ Load-test script
- ✓ Written feedback on your deployment
28 weeks
RM 4,650
payment schedule discussed at enrolment
- ✓ Twice-weekly live sessions
- ✓ Recordings
- ✓ Five artefacts reviewed line by line
- ✓ Capstone review and defence
- ✓ Fortnightly 1:1 office hours
- ✓ Compute credit allowance
- ✓ Cohort channel
- ✓ Repository you own outright
Note: Fees may be claimable under the HRD Corp levy for eligible Malaysian employers. Ask us for details.
[ prompt ]
Not sure which to start with?
Send a message explaining where you are technically. We will read it and tell you which programme, if any, looks like a reasonable next step.