Homework 1 — Your starting point

Due: Tuesday 29 September 2026, 23:59 · folder week-01/ in your shared Drive folder ese-ai-fintech

Goal

Write down where you are starting from, and where you want to be in December — so the course can be shaped around it. Then take the tokenizer apart on text of your own, and put ten of the course’s words into your own language.

This is a lighter week on purpose. The three-asset notebook we would have started today moves to session 2, where we build it together.

Deliverables (in week-01/)

1. README.md — Starting point (400–600 words), four short sections:

   • Background. Two or three projects from your work where data, or its absence, decided something. One that went well, one that did not.
   • What I want to be able to do in December that I cannot do today. Concrete verbs, not topics (“price a pilot’s inference cost”, not “learn about LLMs”).
   • A question I would like answered with data. From your work, your trading, or crypto/DeFi. It does not have to be answerable yet — it is the first draft of your capstone. Say who would use the answer, and what they would decide with it.
   • How I learn best, and one thing that would make this course a waste of your time.

2. tokens.ipynb — the tokenizer cell from the session notebook (§ B1), run on five texts of your choice: at least one in Russian, one with numbers, one company or product name. A table of word count vs token count for each, and a short markdown paragraph: which text surprised you, and what would that mean for the cost of a real project you know?

3. glossary.md — ten terms from the course glossary, at least three from Crypto and DeFi. For each: your definition in one sentence, and one example from your own experience or from a product you know. Where you disagree with the glossary, say so.

4. pitch.md — three sentences from three pitch decks. You will meet a version of each in a data room:

   1. “Our assistant remembers every conversation it has had with your customers.”
   2. “It learns from your corrections in real time.”
   3. “99.2% accuracy on invoice extraction.”

   Two are impossible as stated, one is unanswerable as written. For each: which is it, which part of the machine from today (weights, training, context window, tokens) it is really talking about, and the one question you would ask the vendor. Half a page in total.

Constraints

• Use an assistant as much as you like for the notebook — but every line must be one you can explain. I will ask.
• The README and the glossary are in your words. An assistant may correct the English; it may not write the content.

Scope

3–4 hours. If you are past 4 hours, stop, save what you have, and write down where you got stuck.

Self-check before submitting

• Drive folder ese-ai-fintech exists and is shared with me (see Setup).
• Today’s notebook is saved there, in week-01/.
• tokens.ipynb runs with Runtime ▸ Restart and run all.
• The question in the README names a user and a decision.
• Three of the ten glossary terms are from Crypto and DeFi.
• pitch.md names one question per sentence.

How this feeds the capstone

The question in your README is the first draft of your capstone. We will come back to it in week 3 (can it be framed as a prediction?), week 6 (the one-page proposal), and week 10. It is fine — expected, even — for it to change.