AI and Agents Course

Chapter 1: Foundations of Modern AI

Subtopic	Detail
1.1 Evolution from Classical AI to Foundation Models	Brief history: symbolic AI, statistical ML, deep learning, emergence of foundation & frontier models.
1.2 Key Concepts: Tokens, Embeddings, Parameters	Tokenization strategies, embedding spaces, scaling of parameters and compute laws.
1.3 Model Taxonomy: SLMs vs LLMs vs MLLMs	Small vs large language models, multimodal LLMs, trade-offs in capacity, latency, cost.
1.4 Core Capabilities & Limitations	Reasoning, generation, hallucination, context windows, system prompts, reliability constraints.

Chapter 2: Small Language Models (SLMs)

Subtopic	Detail
2.1 Definition & Roles of SLMs	Edge deployment, on-device inference, privacy and latency benefits.
2.2 Distillation & Quantization	Knowledge distillation, 8/4-bit quantization, QLoRA concepts.
2.3 Popular Open SLM Families	Phi, Mistral small, TinyLlama, Gemma variants—capabilities & benchmarks.
2.4 When to Choose SLMs over LLMs	Cost modeling, latency thresholds, data governance, offline resilience.

Chapter 3: Large Language Models (LLMs)

Subtopic	Detail
3.1 Architecture Recap & Scaling Laws	Transformer blocks, depth vs width, Chinchilla scaling principles.
3.2 Instruction Tuning & Alignment	SFT, RLHF, DPO, safety layers, preference data pipelines.
3.3 Function Calling & Structured Outputs	JSON mode, schema constraints, tool invocation protocols.
3.4 Cost, Latency & Throughput Optimization	Batching, caching, speculative decoding, dynamic routing.

Chapter 4: Multimodal & Vision-Language Models

Subtopic	Detail
4.1 Multimodal Encoders & Fusion	Image, audio, video encoders; joint embedding spaces.
4.2 Vision-Language Architectures	BLIP, Flamingo, LLaVA style adaptors & projection heads.
4.3 Image & Document Understanding	OCR integration, layout parsing, chart/table reasoning.
4.4 Multimodal Tool Use	Grounding outputs with external perception & analysis tools.

Chapter 5: Tool Use & Function Calling

Subtopic	Detail
5.1 Prompt Patterns for Tool Invocation	Schema design, tool selection heuristics, reasoning scaffolds.
5.2 OpenAI / JSON / Schema Approaches	Function calling schemas, JSON schema validation, error recovery.
5.3 Tool Error Handling & Retries	Guardrails, tool feedback loops, self-correction prompts.
5.4 Chaining Tools & Planning	Sequential vs parallel tool execution design patterns (LangChain tool chains, LangGraph stateful graphs).

Chapter 6: Model Context Protocol (MCP) & Open Ecosystems

Subtopic	Detail
6.1 Overview of MCP	Goals: portability, interoperability, decoupling model & tools.
6.2 MCP Server & Client Roles	Resource provision, capability advertisement, session negotiation.
6.3 Integrating MCP with Agents	Routing requests, context expansion, secure tool boundaries.
6.4 Emerging Standards & Protocols	OpenAI tool schemas, LangChain tool interfaces, LangGraph workflow graphs, Open RAG specs.

Chapter 7: Agent Foundations

Subtopic	Detail
7.1 What is an AI Agent?	Definitions: autonomy, goals, environment interaction.
7.2 Core Agent Loop	Perceive → Reason → Act → Reflect cycle.
7.3 Prompt Engineering for Agents	System vs scratchpad vs reflection prompts.
7.4 Memory & State Management	Short-term scratchpads, episodic & semantic memory stores.

Chapter 8: Agent Architectures I

Subtopic	Detail
8.1 ReAct & Chain-of-Thought	Combining reasoning traces with tool/action steps.
8.2 Reflexion & Self-Critique	Iterative refinement loops, critique generation patterns.
8.3 Tree of Thoughts / Graph Search	Exploring reasoning branches (ToT), graph execution with LangGraph, pruning heuristics.
8.4 Multi-Agent Role Specialization	Orchestrators, planners, executors & evaluators.

Chapter 9: Agent Architectures II

Subtopic	Detail
9.1 Hierarchical & Modular Agents	Task decomposition, sub-goal scheduling.
9.2 Planning with LLMs	Backward vs forward planning, dynamic replanning.
9.3 Retrieval-Augmented Agents	Dynamic context injection via vector stores.
9.4 Evaluator & Judge Agents	Quality scoring, safety checks, meta-reasoning roles.

Chapter 10: Retrieval Augmented Generation (RAG)

Subtopic	Detail
10.1 Embeddings & Vector Stores	Embedding models, ANN indexes (FAISS, HNSW).
10.2 Chunking & Index Construction	Semantic vs fixed-size chunking, metadata strategies.
10.3 Retrieval Pipelines	Hybrid search (BM25+vector), rerankers, filtering (LangChain retrievers & LangGraph orchestrations).
10.4 Advanced RAG Patterns	Query rewriting, multi-hop, tool-enriched retrieval.

Chapter 11: Hugging Face & Model Ecosystem

Subtopic	Detail
11.1 Navigating the Hub	Model cards, datasets, spaces, licensing considerations.
11.2 Popular Foundation Models	LLaMA, Mistral, Mixtral, GPT-NeoX, Gemma, T5 families.
11.3 Specialized Models (Vision, Audio)	Whisper, CLIP, Stable Diffusion, Wav2Vec, SAM.
11.4 Fine-Tuning & LoRA Adaptation	PEFT, adapters, parameter efficiency, eval tracking.

Chapter 12: Reasoning & Advanced Prompting

Subtopic	Detail
12.1 Chain-of-Thought & Deliberate	Structured reasoning traces for improved accuracy.
12.2 Program-Aided & Tool-Augmented Reasoning	PAL, code interpreters, symbolic hybrids.
12.3 Self-Consistency & Majority Voting	Ensembles of reasoning paths for reliability.
12.4 Prompt Compression & Context Optimization	Window management, summarization, context distillation.

Chapter 13: Memory & Long-Term Context

Subtopic	Detail
13.1 Episodic vs Semantic Memory	Temporal logs vs knowledge graphs.
13.2 Vector Memory Stores	Persisting conversation embeddings & retrieval strategies.
13.3 Memory Consolidation Patterns	Summarization cycles, forgetting strategies.
13.4 Context Window Extensions	Sliding windows, recurrence, external memory tools.

Chapter 14: Evaluation & Benchmarks

Subtopic	Detail
14.1 Quantitative Benchmarks	MMLU, GSM8K, BBH, MT-Bench: interpreting scores.
14.2 Human & AI Feedback Loops	Preference data, rubric scoring, judge models.
14.3 Agent-Specific Evaluation	Task success, tool efficiency, autonomy metrics.
14.4 Regression & Monitoring Dashboards	Tracking drift, latency, cost, and quality trends.

Chapter 15: Safety, Ethics & Governance

Subtopic	Detail
15.1 Prompt Injection & Jailbreaks	Attack surfaces, sanitization strategies.
15.2 Data Privacy & PII Handling	Redaction, classification, isolation of sensitive content.
15.3 Alignment & Content Moderation	Policy filters, refusal patterns, fallback flows.
15.4 Governance & Auditability	Action logs, reproducibility, compliance frameworks.

Chapter 16: Deployment & Infrastructure

Subtopic	Detail
16.1 Serving Architectures	Gateway, model router, scaling replicas, autoscaling.
16.2 Caching & Session Strategies	KV caches, prefix reuse, partial decoding.
16.3 Observability & Telemetry	Tracing prompts, metrics, structured logging.
16.4 Cost Optimization Techniques	Model routing, tiered quality, re-ranking cascades.

Chapter 17: Optimization & Performance Engineering

Subtopic	Detail
17.1 Inference Optimization	Quantization, pruning, speculative & streaming decode.
17.2 Hardware Acceleration	GPUs vs TPUs vs NPUs; attention offloading.
17.3 Throughput & Concurrency Control	Admission control, batching algorithms.
17.4 Latency Diagnostics	Token breakdown, network vs compute bottlenecks.

Chapter 18: Specialized Agent Modalities

Subtopic	Detail
18.1 Code Agents & Execution Sandboxes	Execution safety, timeouts, state isolation.
18.2 Data Analysis & Notebook Agents	EDA automation, visualization toolchains.
18.3 Robotic & Embodied Agents	Perception-action loops, simulation-to-real gaps.
18.4 Autonomous Web & API Agents	Browser automation, DOM plans, API chaining.

Chapter 19: Future Directions & Research Frontiers

Subtopic	Detail
19.1 Continual & Lifelong Learning	Adaptive fine-tuning, catastrophic forgetting mitigation.
19.2 Neuro-Symbolic & Hybrid Systems	Symbol grounding, reasoning augmentation.
19.3 Open Weights vs Closed Models	Economics, innovation velocity, community impact.
19.4 Autonomous Alignment & Self-Evaluation	Agentic red teaming, reflective safety loops.

Chapter 20: Capstone & Integration

Subtopic	Detail
20.1 Designing an End-to-End Agent System	Blueprint: ingestion, memory, planning, tool layer.
20.2 Production Hardening Checklist	Resilience, fallbacks, rate limiting, observability.
20.3 Case Studies & Reference Architectures	Patterns across customer support, analytics, automation.
20.4 Next Steps & Learning Paths	Open research problems, community resources, benchmarking.

Chapter 21: Practical LangChain/LangGraph Implementation Patterns

Subtopic	Detail
21.1 LangChain Fundamentals & Core Components	Chains, prompts, memory, retrievers, agents - building blocks and composition patterns.
21.2 LangGraph State Machines & Workflow Design	Graph-based agent flows, state management, conditional routing, human-in-the-loop patterns.
21.3 Advanced Integration Patterns	Custom tools, multi-model routing, error handling, streaming responses, production deployment.
21.4 Performance Optimization & Debugging	LangSmith tracing, async execution, caching strategies, cost optimization techniques.

Start Course →