A study companion that lives on the macOS desktop, with chat, word lookup, live captions and notes after class. This post covers the scope decisions, the model comparisons and a few things I learned along the way.
AI Agents, From Chatbot to Production An AI agent is the LLM application taken further — something that can plan on its own, carry out tasks, and interact with its environment. Foundational papers: Yao, S., et al. (2022). “ReAct: Synergizing Reasoning and Acting in Language Models.” ICLR. [paper] — the ReAct framework, the core pattern behind most agents Wei, J., et al. (2022). “Chain-of-Thought Prompting Elicits Reasoning in Large Language Models.” NeurIPS. [paper] Wang, G., et al. (2024). “A Survey on Large Language Model based Autonomous Agents.” arXiv. [paper] — a survey of the agent landscape 1. The Overall Progression Chatbot → Tool Use → Agent → Production Agent What each stage means: ...
Prompt Engineering https://www.coursera.org/learn/chatgpt-prompt-engineering-for-developers-project/home/welcome Prompt engineering is the practice of designing and refining the input you give a large language model so it produces the output you actually want. It took off after GPT-3 came out, and it’s basically the bridge between what a user wants and what the model can do. Key papers: Brown, T. B., et al. (2020). “Language Models are Few-Shot Learners.” NeurIPS. [link] — the GPT-3 paper, the first to really demonstrate the potential of prompt engineering Wei, J., et al. (2022). “Chain-of-Thought Prompting Elicits Reasoning in Large Language Models.” NeurIPS. [link] 1. Prompting Principles Principle 1: Write clear and specific instructions Clarity and specificity are the two things that matter most when writing a prompt. Vague instructions tend to produce unpredictable outputs, while specific ones noticeably improve output quality. ...
RAG RAG (Retrieval-Augmented Generation) combines information retrieval with text generation — it pulls in an external knowledge base to make an LLM’s answers better. Foundational papers: Lewis, P., et al. (2020). “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks.” NeurIPS. [paper] — the paper that started RAG Guu, K., et al. (2020). “REALM: Retrieval-Augmented Language Model Pre-Training.” ICML. [paper] Izacard, G., & Grave, É. (2021). “Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering.” EACL. [paper] — a different take on RAG 1. Context Window What it is The context window is the maximum amount of text an LLM can read and reason over in a single pass. ...
Transformer: From First Principles to a Product Lens 📚 Why does the Transformer matter so much? The Transformer architecture came out of Vaswani et al.’s 2017 paper “Attention Is All You Need.” It changed NLP completely and is now the backbone behind every major LLM — GPT, BERT, LLaMA, you name it. The core idea: drop CNNs and RNNs entirely and rely only on the attention mechanism to capture dependencies across a sequence. That single move solved two problems at once: Parallelism: RNNs process tokens one at a time; attention lets you process the whole sequence at once, which speeds up training enormously Long-range dependencies: RNNs/LSTMs tend to lose the signal over long sequences (vanishing gradients); attention doesn’t have that problem ...
1. Comparing the mainstream payment methods 1. Business-to-business cash transfer How it works: an initiated transfer from a bank counter or corporate online banking, or a payment through a bank’s fund-splitting system with a receipt. Limits: large transfers are possible. Anything above RMB 50,000 goes through the PBOC’s large-value payment system, which doesn’t run on weekends and holidays, and can be affected accordingly (on weekdays it runs from 20:30 the previous evening through 17:15 the same day). ...
Notes from working through Cross-Border E-Commerce Payment and Settlement, my starting point for getting into cross-border payments. A few basic concepts Foreign exchange (FX) Foreign exchange is a claim held by a country’s monetary authorities — the central bank, currency management bodies, exchange stabilization funds, and the treasury — in the form of bank deposits, treasury bills, and short- and long-term government securities, that can be drawn on to cover a balance-of-payments deficit. It covers foreign currency itself, foreign-currency deposits, foreign-currency securities (government bonds, treasury bills, corporate bonds, stocks), and foreign-currency payment instruments such as drafts, bank certificates of deposit, and postal savings certificates. ...
Wrapping up what to prepare for when you’re new to a job, and the mistakes I made so I don’t forget them. 1. Get your tools set up Find out what software the company runs on day to day. If IT already has it installed for you, great. If it’s something you’ve never used, learn the basics when you have a spare moment, or just ask a colleague. 2. Learn how the place actually works ...
When you’re digging into what users actually need, it’s rarely something concrete and explicit. There’s the old Ford line — “if I’d asked people what they wanted, they’d have said a faster horse” — but once you get to actually building the product, you still have to land on something concrete: specific features that satisfy the underlying need. So what I mean by “requirement analysis and management” here is the whole chain: start from a surface-level ask, dig into the real need underneath it, evaluate it, land on a specific solution, and hand that off to engineers and designers to turn into an actual product. B2B and B2C products differ quite a bit in how this plays out. ...
1 Why Bother with Competitive Analysis 1. Better decisions. Understanding competitors’ products helps you position your own — and steer clear of the strongest players so you don’t wander into a red ocean. It lowers the risk of getting your bets wrong. 2. Learning from others. Studying what makes a competitor’s product strong lets you build more deliberately: take what works, absorb their strengths, and use it to sharpen your own edge. ...