FDE AI Factory roadmap
The entire production-AI journey is visible from day one. Free lessons are complete anchor topics across the path; Pro unlocks the surrounding sequence, practice, and build-ready depth.
Free
Read the public library and complete selected anchor lessons across each track.
Pro
Follow the full self-paced sequence, exercises, and production implementation path.
Live Cohort
Add instructor feedback, reviews, deadlines, and a reviewed capstone.
Open knowledge library
The Knowledge Lab stays public: use it to understand the systems before you decide whether to follow the structured Pro path.
Tokens, embeddings, prompting, inference, agents, and evaluation.
Vector search, similarity, kNN, sparse retrieval, hybrid search, and quality.
From simple retrieval and chunking to reliable, production-ready RAG patterns.
Graphs, ontology, GraphRAG, and connected context for AI systems.
Structured learning paths
These are full lessons—not teaser pages. Each shows a meaningful part of the path; the locked topics around them form the complete Pro sequence.
The math operations behind embeddings, retrieval, training, and attention.
Similarity With Dot Product
The one operation behind embeddings, search, and attention.
Softmax Probabilities
Turn raw scores into probabilities that sum to one.
Tensor Broadcasting
How arrays of different shapes do math together.
L1 vs L2 Norms
Two ways to measure the size of a vector.
Matrix Multiplication as Transformation
How a matrix reshapes, rotates, and scales a vector.
Cosine vs Euclidean
When direction matters, and when distance does.
Gradients & Gradient Descent
How models learn by following the slope downhill.
Logits to Probabilities
From raw scores to probabilities and log-probs.
Vectors & Vector Spaces
The arithmetic behind king − man + woman = queen.
Attention as a Weighted Average
Softmax the scores, then blend the values.
Build the architecture component by component, from tokens through inference.
Tokenization (BPE)
Raw text → integer token IDs via Byte-Pair Encoding
Token Embeddings
One-hot × Wₑ × √d_model — sparse IDs to dense vectors
Positional Encoding
Sinusoidal functions that inject sequence order
Scaled Dot-Product Attention
softmax(QKᵀ/√d_k)·V — the core operation
Multi-Head Attention
h parallel heads, same cost, richer representation
Causal Masking
Blocking future tokens with −∞ before softmax
Layer Normalization
γ⊙(x−μ)/√(σ²+ε)+β — stabilizing activations
Feed-Forward Network
max(0, xW₁+b₁)W₂+b₂ — the per-position nonlinearity
Residual Connections
LayerNorm(x + Sublayer(x)) — the gradient highway
Encoder–Decoder & Cross-Attention
The full architecture and three attention types
Training
Cross-entropy, label smoothing, warmup schedule, dropout
Inference & Generation
Beam search, temperature, KV-cache, sampling
End-to-End Trace
Complete forward pass — every tensor, every multiply
Live Playground
Real embeddings — type text, see vectors
Pro path previews
The articles explain the ideas openly. Pro connects them into a complete, guided implementation path with practice, evaluation, and your own production decisions.
Production search
Move beyond brute-force kNN: build an HNSW index, measure Recall@10 against exact ground truth, and tune ef_search to meet the real latency target.
Open HNSW tuning workspaceProduction RAG
Build the full flow: prepare documents, retrieve the best context, generate a grounded response, and measure the result before shipping.
Graph retrieval
Extract entities and relationships, traverse the relevant connections, merge graph and vector context, then generate a multi-hop answer.
Complete the path
Unlock the complete self-paced sequence, then use the production search content to apply those ideas to your own systems. Live Cohort adds feedback and capstone review.