How to build · 10

AI Development

LLM APIs, prompting, AI coding assistants, embeddings, RAG, tool calling, agents, evaluation and security.

Read: 0 / 16AI Fundamentals LLM APIs Prompt Engineering AI-Assisted Coding Embeddings & Vector Databases RAG Tool Calling AI Agents & Automation AI Evaluation AI Security Vibe Coding

AI Fundamentals

Fundamentals✓
LLMs: how language models work and what they can do
A large language model (LLM) is a neural network trained to predict the next piece of text (a token) from everything that came before it; question answering, translation, classification and code generation all grow out of that skill. The model only sees what fits in its context window, remembers nothing between requests and can be confidently wrong, so in a product it is backed by data, format validation and human review.
Beginner · 9 min read

LLM APIs

Tutorial✓
LLM APIs: a first request to Claude and OpenAI
To call a language model from code, install the official SDK (@anthropic-ai/sdk or openai), keep the key in a server-side environment variable and send a request: at Anthropic that is client.messages.create with model, max_tokens, system and messages; at OpenAI it is client.responses.create with model, instructions and input. Interfaces use streaming, and the SDKs retry 429 and 5xx errors on their own by default; your job is to set limits and a timeout and handle the remaining errors by type.
Beginner · 7 min read · practice ~1.5–3 h

Prompt Engineering

How-to✓
Prompt engineering: instructions, examples, output structure
A good prompt is a specification: a role, the context and the reason for the task, step-by-step rules, 3–5 diverse examples and an exact output format, with variable data separated by XML tags. For machine processing, get the answer as structured output constrained by a JSON schema (output_config.format at Anthropic, text.format at OpenAI) and still validate it in code, for example with a Zod schema. Measure prompt quality on a labeled set after every change.
Beginner · 7 min read · practice ~30–60 min

AI-Assisted Coding

How-to✓
AI assistants in development: Claude Code and Codex in practice
Claude Code and Codex are agentic tools: they read the repository, edit files and run commands on their own instead of completing a single line. To make an agent helpful rather than harmful you need three things: a project instruction file (AGENTS.md or CLAUDE.md) with commands and restrictions, permissions and a sandbox that technically limit what it can do, and a task with verifiable acceptance criteria. The agent's output goes through the same tests and review as human code.
Intermediate · 11 min read · practice ~30–60 min

Embeddings & Vector Databases

Tutorial✓
Embeddings and vector search with pgvector
An embedding is a vector of several hundred numbers that a model computes from a text so that texts with similar meaning get nearby vectors. You can store the vectors directly in PostgreSQL with the pgvector extension, find the nearest ones by cosine distance with the <=> operator, and add an approximate HNSW or IVFFlat index for large tables. Embeddings can come from a local model with no API keys or from a cloud API — what matters is encoding documents and queries with the same model.
Intermediate · 11 min read · practice ~1.5–3 h

RAG

Tutorial✓
RAG: answers grounded in your documents
RAG (retrieval-augmented generation) is a setup where, before the model answers, your program finds a few relevant chunks in a knowledge base and passes them into the prompt; the model answers only from them and cites the source. The pipeline has an indexing phase (chunking, embeddings, a vector database) and an answering phase (top-k retrieval, context assembly, generation with citations). Quality is checked separately: retrieval with recall@k on test questions, the answer by checking citations and an honest “don't know”.
Intermediate · 11 min read · practice ~3–6 h

Tool Calling

Tutorial✓
Tool calling: the model calls your functions
Tool calling is a mechanism where you describe functions (name, purpose, JSON Schema of arguments), the model replies with a structured call request (tool_use), your code runs the function and returns the result (tool_result), and the model continues its answer. The model executes nothing itself: validate arguments in code, return failures with is_error: true, and confirm actions with consequences outside the model.
Intermediate · 11 min read · practice ~1.5–3 h

AI Agents & Automation

Deep dive✓
AI agents: the loop, tools, boundaries and automation
An AI agent is a language model that picks tools and next steps in a loop until it reaches a goal, getting feedback from its environment after every action. Unlike a workflow, where code sets the order of steps, the agent decides for itself, so it needs step and budget limits, a narrow tool set, human approval for actions with consequences, and a step journal. Start with the simplest solution: one model call or a fixed workflow is often enough.
Advanced · 11 min read

AI Evaluation

How-to✓
Evaluating LLM features: test sets and metrics
You measure an LLM feature with an eval set: dozens of real inputs with reference answers that run through automated checks after every prompt or model change — exact match and schema and rule validation in code, and, where meaning matters, an LLM judge with a rubric that has been checked against human labels. The output is a report with pass rates per check and per slice, a list of regressions, cost and latency, and in CI it can stop a release.
Advanced · 15 min read · practice ~1–2 h

AI Security

Fundamentals✓
LLM application security: prompt injection and data leaks
An LLM application gains a new channel of untrusted input: the model follows instructions from any text in its context — from the user, a document, an email or a web page — and cannot reliably tell them apart from data. So the defence is built around the model rather than in the prompt: instructions are separated from data, tools get minimal permissions, a person confirms actions with consequences, and deterministic code checks the model's output against a schema and an allowlist of actions. The baseline is the 2025 edition of the OWASP Top 10 for LLM Applications.
Intermediate · 12 min read

Vibe Coding

Learning paths with this direction

AI integrations · 8 →Vibe coding: from prototype to product · 7 →← All directions

Tell us about your task — we'll review your request and get back to you

Got a project?
We'll scope and estimate it — for free.

Tell us about the product or the problem. After reviewing the task we'll come back with an architecture, a timeline and a price for the first version.

Scope and risks estimated before we start
NDA — yours or ours
Work starts within a week

Message on WhatsApp