🧪 The Lab
Dig to the bottom of each topic and see it work. You write the implementation; the Lab makes the internals visible, runnable, and tweakable — so you can prove you truly understand it and explain it to anyone.
AI engineering
— bounded numeric and mock-model engines with explicit evidence labelsedit a bounded fixture → calculate Q/K/V, attention, logits, and seeded sampling
prompt → tokens → greedy vs sampled, temperature/seed
model emits a call → host runs it → result fed back
RAG, ReAct, memory, MCP, multi-agent, ops
Run real code
— browser runtimes and bounded server fixtures — each route states what actually ranreal CPython in your browser (WASM) — write & run, no server
live Sandpack runner in the Dojo React lessons
edit one reviewed source → run a fixed Java 21 child JVM → replay classfile, main, reference-value copy, and GC unknowns
Data structures
— code it yourself, watch it moveOpenJDK 21 semantic model: lazy table, collisions, resize split, treeify, mutable-key failure
capacity, size, amortized doubling
LIFO, one end, O(1)
ring buffer, head/tail wrap
nodes, pointers, reversal
parent[], path compression, rank
immutability, interning, char vs byte
BST insert, rotation, traversal
BFS/DFS frontier, adjacency
sift up/down, heapify
prefix paths, node explosion
Algorithms
— specification → invariant → counterexample → proof transferSystems & reliability
— watch the dynamics, not just the diagramdedup key, retry-safe writes
type transition rules as code → graph redraws live
backoff, jitter, poison → DLQ
cache-aside, TTL, stampede, hot key
partitions, offsets, rebalance, lag
predict B-tree key order → real EXPLAIN ANALYZE → defend plan and counterexample
resources, idempotency, versioning
follow one request through Spring, Hikari, JDBC, and PostgreSQL — then move its work
Production pitfalls & load
— watch it break as traffic grows, then fix itedit a real patch → run two consumer workloads → prove context reversal
real Hikari active/waiting/timeout captured replay
real PostgreSQL 40P01 victim + committed survivor
review source → predict proxy advice → inspect real post-failure rows
hot key expires → N requests hammer DB
edit one Spring path → replay two real transactions → preserve a patch ledger
fixed window edge, token bucket
produce > consume → lag grows
at-least-once → dedup key
API paradigms
— really understand what each ISresources, verbs, status, statelessness
protobuf, HTTP/2 streams, codegen
one graph, client-picked fields, N+1
Every route states whether its evidence is measured, calculated, modeled, or illustrative. Some slices complete prediction → change → evidence → defense/transfer; an interactive model alone is not presented as mastery proof.