Who This Track Is For
Companies that want Lead AI Engineers who can:
Build LLM applications
RAG, Agents
Understand model internals
Fine-tuning, evaluation
Make architecture decisions
When to fine-tune vs prompt
Lead ML/AI teams
Technical credibility
Bridge product & research
Cross-functional collaboration
Target Companies
| Type | Examples | Why ML Matters |
|---|---|---|
| Big Tech | Google, Apple, Amazon | Expect ML depth |
| AI-First Startups | Mistral, Aleph Alpha | Building models |
| Automotive AI | BMW, Bosch | Custom models for edge |
| Research-Adjacent | DeepMind, Anthropic | ML fundamentals required |
Skills Matrix
Core (Must Have)
Python for ML
Critical
RAG Systems
Critical
LLM APIs & Prompting
Critical
PyTorch Basics
High
Fine-tuning (LoRA)
High
AI Agents
High
LLM Evals
High
Supporting (Good to Have)
Transformers Architecture
Medium
Training Basics
Medium
MLOps
Medium
Classical ML
Low
6-Week Curriculum
1
Python + ML Foundations
Solid Python for ML, understand how models work
Day 1-2: Python for ML (10 hours)
- NumPy (arrays, broadcasting)
- Pandas (DataFrames)
- Async/await
- Type hints + Pydantic
Day 3-4: PyTorch Fundamentals (10 hours)
import torch
import torch.nn as nn
# Tensors
x = torch.tensor([1, 2, 3])
y = x.cuda() # GPU
# Autograd
x = torch.tensor([2.0], requires_grad=True)
y = x ** 2
y.backward()
print(x.grad) # 4.0
# Neural Network
class SimpleNet(nn.Module):
def __init__(self):
super().__init__()
self.fc = nn.Linear(10, 1)
def forward(self, x):
return self.fc(x)Day 5-6: Transformers Architecture (8 hours)
Understand conceptually:
- Self-attention mechanism
- Multi-head attention
- Positional encoding
- Encoder vs Decoder
- Why transformers scale
Day 7: LLM APIs + Prompting (6 hours)
Set up OpenAI, Anthropic APIs. Practice 10+ prompt patterns.
Week 1 Deliverables
- [ ] PyTorch basics working
- [ ] Trained simple model (MNIST)
- [ ] Can explain transformer architecture
- [ ] LLM APIs set up
2
RAG Systems + Embeddings Deep Dive
Build RAG with understanding of how embeddings work
Day 1-2: Embeddings Theory + Practice (8 hours)
- Word embeddings (Word2Vec concept)
- Sentence embeddings (how they're trained)
- Contrastive learning basics
- Embedding dimensions and similarity
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
model = SentenceTransformer('all-MiniLM-L6-v2')
embeddings = model.encode(["This is a test", "Another sentence"])
sim = cosine_similarity([embeddings[0]], [embeddings[1]])Day 3-4: Vector Databases + Retrieval (8 hours)
Understand: HNSW algorithm, ANN vs exact search, metadata filtering, hybrid search
Day 5-6: LangChain RAG (8 hours)
from langchain.text_splitter import RecursiveCharacterTextSplitter
# Understand WHY these parameters
splitter = RecursiveCharacterTextSplitter(
chunk_size=1000, # Why 1000?
chunk_overlap=200, # Why overlap?
separators=["\n\n", "\n", " ", ""] # Priority order
)Week 2 Deliverables
- [ ] Understand how embeddings work (can explain)
- [ ] RAG with multiple retrieval strategies
- [ ] Benchmark different chunking approaches
3
Fine-tuning + AI Agents
Know when and how to fine-tune, build agents
Day 1-2: Fine-tuning Fundamentals (10 hours)
| Use Case | Prompting | Fine-tuning |
|---|---|---|
| General tasks | Yes | No |
| Domain terminology | Maybe | Yes |
| Cost optimization | No | Yes (smaller model) |
# LoRA: Low-Rank Adaptation
# Instead of updating all weights W, learn small matrices A and B
# W' = W + BA where B is (d x r) and A is (r x k), r << d
from peft import LoraConfig, get_peft_model
config = LoraConfig(
r=8, # Rank (smaller = fewer params)
lora_alpha=32, # Scaling factor
target_modules=["q_proj", "v_proj"],
lora_dropout=0.1,
)
# Only ~0.1% of parameters are trainable!Day 3-4: LangChain Agents (8 hours)
Build agents with tools: search, calculator, database queries.
Day 5-6: LangGraph Workflows (8 hours)
Stateful agents with planning, execution, review nodes.
Week 3 Deliverables
- [ ] Fine-tuned small model on custom task
- [ ] Understand when fine-tuning makes sense
- [ ] Agent with 3+ tools
- [ ] LangGraph workflow with state
4
Evaluation + ML System Design
Measure quality, design ML systems
Day 1-2: LLM Evaluation (8 hours)
RAGAS for RAG, DeepEval for general LLM evaluation.
Day 3-4: ML System Design (10 hours)
User Query
|
v
+------------------+
| Query Analysis | <-- Intent, entities, query rewrite
+--------+---------+
|
+----+----+
v v
+-------+ +-------+
|Vector | |Keyword| <-- Hybrid retrieval
|Search | |Search |
+---+---+ +---+---+
| |
+----+----+
v
+------------------+
| Reranking | <-- Cross-encoder
+--------+---------+
v
+------------------+
| LLM Generation | <-- With context
+--------+---------+
v
+------------------+
| Evaluation | <-- Faithfulness check
+------------------+
Practice Questions
- "Design a document search system for legal documents"
- "Design a chatbot with RAG for customer support"
- "Design a recommendation system using embeddings"
Week 4 Deliverables
- [ ] Evaluation pipeline for RAG
- [ ] ML system design practice (3 designs)
- [ ] Production API with monitoring
- [ ] Can explain trade-offs in interviews
5
Portfolio + Advanced ML Topics
Polish portfolio, learn advanced topics for interviews
Day 1-2: Advanced ML Concepts (8 hours)
Attention Mechanism
Attention(Q,K,V) = softmax(QK^T / sqrt(d_k)) V
Loss Functions
Cross-entropy, MSE, Contrastive
Optimization
Adam, LR scheduling, Mixed precision
Scaling
Data/Model parallelism, Checkpointing
Day 3-5: Capstone Project (15 hours)
Option A
Domain-Specific RAG + Fine-tuning
Option B
Multi-Agent Research System
Option C
ML Pipeline Demo
Week 5 Deliverables
- [ ] Capstone project complete
- [ ] 3 portfolio projects
- [ ] Blog posts published
- [ ] Demo videos
6
Job Search + Interview Prep
Apply and prepare for ML-depth interviews
Interview Questions Bank
ML Fundamentals
- 1. "Explain how transformers work"
- 2. "How does LoRA reduce training cost?"
- 3. "When would you fine-tune vs use RAG?"
System Design
- 1. "Design a RAG system for customer support"
- 2. "Design an ML serving system"
Week 6 Deliverables
- [ ] 25+ applications
- [ ] Interview prep complete
- [ ] Mock interviews done
- [ ] Network connections made
Success Metrics
By Week 6, You Should:
Can explain transformers architecture
Can implement RAG from scratch
Understand when to fine-tune vs RAG
Built evaluation pipeline
3 portfolio projects with ML depth
Confident in ML system design
Ready to Start?
Choose your track and begin your journey