This is the AI LaunchPad track syllabus — 6 months · 24 weeks.View the full program
What you'll be able to do
- Call LLM APIs and write async Python that runs LLM calls concurrently and streams responses
- Engineer prompts and force validated, structured (Pydantic) outputs from models
- Build RAG systems with embeddings, ChromaDB / pgvector, chunking, and re-ranking
- Master LangChain and LangGraph to build stateful, multi-step, multi-agent systems
- Integrate tools and data into agents with MCP (Model Context Protocol)
- Serve agents as production APIs with FastAPI, containerize with Docker, and deploy to the cloud
- Apply production AI engineering: evaluation, observability/tracing, and reliability
Tools & technologies
REST APIshttpxAsync PythonPydanticPrompt EngineeringStructured OutputsStreamlitEmbeddingsVector DatabasesChromaDBpgvectorChunking
The syllabus
Python Programming and Logic Building
Python and Programming Fundamentals
- Day 1AI, Generative AI and LLM career paths · Python, VS Code and Jupyter setup · First Python program and notebook workflowBuild: GenAI Learning Starter Notebook
- Day 2Variables and identifiers · Numbers, strings and Boolean values · Input, output and type conversionBuild: Interactive User Profile Generator
- Logic Building 1Problem decomposition · Algorithms and pseudocode · Flowcharts and sequence-based problemsBuild: AI Application Workflow Designer
- Day 4Arithmetic and assignment operators · String indexing and slicing · String methods and formatted outputBuild: Smart Text Formatter
- Day 5Comparison and logical operators · if, elif and else · Nested conditions and validationBuild: Rule-Based Content Classifier
- Logic Building 2Decision tables · Multi-condition problem solving · Boundary cases and branch debuggingBuild: User Query Routing Engine
Loops, Collections and Advanced Logic
- Day 7while loops · Counters, accumulators and sentinel values · Loop-based input validationBuild: Conversational Menu Simulator
- Day 8for loops and range · Nested loops · break, continue and iteration controlBuild: Text Statistics Generator
- Logic Building 3Searching and counting problems · Pattern and sequence logic · String-processing problemsBuild: Python Logic Challenge Toolkit
- Day 10Lists and tuples · Indexing, slicing and collection methods · List comprehensionsBuild: Prompt Library Manager
- Day 11Dictionaries and sets · Frequency counting · Membership and duplicate detectionBuild: Word-Frequency Analyser
- Logic Building 4Collection-based algorithms · Searching and similarity rules · Menu-driven application designBuild: Rule-Based FAQ Matcher
Functions, Files, OOP and Development Workflow
- Day 13Functions and parameters · Return values and scope · Lambda functions and reusable utilitiesBuild: Text Processing Utility Library
- Day 14Text, CSV and JSON files · Exceptions and validation · Logging and debuggingBuild: Fault-Tolerant Dataset Loader
- Day 15Classes and objects · Dataclasses and composition · Modules, packages and virtual environmentsBuild: Prompt Template Management System
- Day 16Git and GitHub · Branches, commits and pull requests · README, requirements and repository structureBuild: Version-Controlled GenAI Repository
- Day 17NumPy array foundations · pandas DataFrames · Dataset inspection and basic transformationsBuild: Text Dataset Profiler
- Day 18Multi-file Python architecture · File, OOP and dataset integration · Testing and documentationBuild: CLI Dataset Preparation Tool
AI, Mathematics and Machine-Learning Foundations
Foundations Required for GenAI
- Day 19Artificial intelligence, ML, DL and GenAI · Supervised, unsupervised and generative learning · AI problem framing and success metricsBuild: GenAI Problem-Framing Canvas
- Day 20Scalars, vectors and matrices · Dot products and matrix multiplication · Norms, distances and cosine similarityBuild: Vector Similarity Search Engine
- Day 21Probability and random variables · Probability distributions · Logits, probabilities and softmaxBuild: Token Probability Simulator
- Day 22Features, labels and datasets · Training, validation and test sets · Baselines and data leakageBuild: Leakage-Safe Dataset Splitter
- Day 23Classification metrics · Overfitting and underfitting · Regularisation and error analysisBuild: Model Evaluation Report
- Day 24Scikit-learn pipeline · Text vectorisation · Baseline model training and comparisonBuild: Message Intent Baseline Classifier
PyTorch and Deep-Learning Foundations
Neural Networks for Language Modelling
- Day 25PyTorch tensors and shapes · Tensor operations and devices · Automatic differentiationBuild: PyTorch Tensor Laboratory
- Day 26Datasets and DataLoaders · Batching and shuffling · CPU and GPU workflowBuild: Custom Text DataLoader
- Day 27Artificial neurons and layers · Activation functions · Loss functions and optimisersBuild: Neural Binary Classifier
- Day 28Forward and backward propagation · Training and validation loops · Checkpoints and reproducibilityBuild: Reusable PyTorch Training Engine
- Day 29Embedding layers · Sequence batching and padding · RNN, LSTM and GRU foundationsBuild: Sequence Sentiment Classifier
- Day 30Autoencoders and GANs · Autoregressive and diffusion models · Generative-model selection criteriaBuild: Generative Architecture Comparison Studio
Classical Natural Language Processing
NLP Before Transformers
- Day 31Text corpora and documents · Unicode and text normalisation · Regular expressions and text cleaningBuild: Reusable Text Cleaning Pipeline
- Day 32Sentence and word tokenisation · Stop-word removal · Stemming and lemmatisationBuild: NLP Preprocessing Toolkit
- Day 33N-grams · Bag of Words · TF-IDF and sparse matricesBuild: TF-IDF Document Search Engine
- Day 34Text-classification workflow · Naive Bayes and logistic regression · Classification evaluationBuild: Support-Ticket Classifier
- Day 35Word embeddings · Word2Vec and GloVe concepts · Semantic similarity and analogyBuild: Semantic Job-Matching Tool
- Day 36Part-of-speech tagging · Named Entity Recognition · Topic modelling and NLP error analysisBuild: Resume Entity Extraction System
Transformer Models
Transformer Architecture from First Principles
- Day 37NLP and large language models · Limitations of RNNs and LSTMs · Transformer capabilities and use casesBuild: Transformer Use-Case Mapper
- Day 38Attention intuition · Query, key and value representations · Attention-score calculationBuild: Attention Weight Visualiser
- Day 39Scaled dot-product attention · Self-attention · Padding and causal masksBuild: From-Scratch Self-Attention Layer
- Day 40Multi-head attention · Positional encoding · Feed-forward layers, residuals and normalisationBuild: Transformer Block Builder
- Day 41Encoder-only architecture · Decoder-only architecture · Encoder-decoder architectureBuild: Transformer Architecture Selector
- Day 42Masked-language modelling · Causal-language modelling · Sequence-to-sequence trainingBuild: Tiny Transformer Text Classifier
Hugging Face Transformers
Using Pretrained Transformer Models
- Day 43Hugging Face ecosystem · Model Hub and checkpoints · Transformer pipelinesBuild: Multi-Task Transformer Explorer
- Day 44AutoTokenizer and AutoModel classes · Model configuration · Token IDs, masks and special tokensBuild: Transformer Input Inspector
- Day 45Text classification · Token classification and NER · Extractive question answeringBuild: Transformer NLP Task Suite
- Day 46Text-generation pipeline · Greedy and beam decoding · Temperature, top-k and top-p samplingBuild: Controlled Text Generation Lab
- Day 47Summarisation · Translation · Encoder-decoder inferenceBuild: Multilingual Summarisation Tool
- Day 48Batching and device placement · Context limits and model selection · Bias, hallucinations and model limitationsBuild: Transformer Inference Studio
Datasets, Tokenizers and Model Sharing
- Day 49Hugging Face Datasets library · Dataset loading and inspection · Dataset features and splitsBuild: Public NLP Dataset Explorer
- Day 50map, filter and select operations · Batched transformations · Caching and train-validation splittingBuild: Dataset Transformation Pipeline
- Day 51Tokenizer normalisation · Pre-tokenisation · BPE, WordPiece and Unigram tokenisationBuild: Custom Subword Tokenizer
- Day 52Data collators · Dynamic padding and truncation · Metrics and evaluation datasetsBuild: Transformer Batch Preparation Engine
- Day 53Model and tokenizer repositories · Model cards and dataset cards · Versioning, licensing and model sharingBuild: Published Model Card Repository
- Day 54Dataset curation · Custom tokenizer training · Reusable preprocessing pipelineBuild: Indic-Language Dataset and Tokenizer Pipeline
LLM Inference and Prompt Engineering
Building Reliable LLM Interactions
- Day 55LLM inference architecture · Hosted and local-model inference · Tokens, context and response lifecycleBuild: Local and Hosted Model Comparator
- Day 56Generation parameters · Deterministic and stochastic generation · Streaming and stopping conditionsBuild: LLM Generation Control Panel
- Day 57Prompt anatomy · Zero-shot and few-shot prompting · Roles, context, examples and constraintsBuild: Reusable Prompt Template Library
- Day 58Task decomposition · Reasoning-oriented prompting patterns · Self-consistency and response verificationBuild: Stepwise Problem-Solving Assistant
- Day 59Structured outputs · JSON and schema-constrained responses · Tool and function callingBuild: Structured Information Extractor
- Day 60Prompt evaluation · Prompt injection and unsafe instructions · Output validation and guardrailsBuild: Secure Customer-Support Copilot
Retrieval-Augmented Generation
RAG and Vector Databases
- Day 61RAG architecture · Parametric and external knowledge · Prompting, RAG and fine-tuning decisionsBuild: RAG Solution Design Canvas
- Day 62PDF, text, web and CSV loaders · Document parsing · Metadata extraction and document IDsBuild: Multi-Source Document Ingestion System
- Day 63Character and token chunking · Semantic and recursive splitting · Chunk overlap and metadata preservationBuild: Chunking Strategy Benchmark
- Day 64Text embeddings · FAISS and Chroma vector stores · Similarity search and persistenceBuild: Semantic Document Search Engine
- Day 65Similarity and MMR retrieval · Hybrid search and reranking · Query expansion and rewritingBuild: Advanced Retrieval Pipeline
- Day 66Retrieval and generation integration · Source-grounded answers · Retrieval and response evaluationBuild: Document Question-Answering RAG Application
LangChain and Agentic Applications
Chains, Tools and Agents
- Day 67LangChain architecture · Chat models and prompt templates · Output parsers and runnable componentsBuild: LangChain Component Laboratory
- Day 68Sequential and parallel chains · Batch and streaming execution · Retries, fallbacks and configurable modelsBuild: Resilient LLM Processing Chain
- Day 69Conversation history · History-aware retrievers · Memory and context-window managementBuild: Conversational RAG Chatbot
- Day 70Tools and tool schemas · Agent loops · Structured tool selection and executionBuild: Calculator and Search Agent
- Day 71SQL toolkit · Website and search tools · Agent routing and tool permissionsBuild: Natural-Language SQL Assistant
- Day 72Agentic RAG · Multi-tool workflow · Agent evaluation and failure handlingBuild: Research and Knowledge Assistant
Fine-Tuning Pretrained Models and LLMs
Model Adaptation
- Day 73Prompting versus RAG versus fine-tuning · Transfer learning and domain adaptation · Fine-tuning requirements and cost decisionsBuild: Model Adaptation Decision Assistant
- Day 74Task-specific dataset preparation · Label mapping and tokenisation · Transformer classification fine-tuningBuild: Fine-Tuned Intent Classifier
- Day 75Causal language-model datasets · Instruction and chat templates · Supervised fine-tuning workflowBuild: Instruction Dataset Formatter
- Day 76Parameter-efficient fine-tuning · LoRA and adapter layers · QLoRA and quantised trainingBuild: LoRA Domain Adapter
- Day 77Training arguments and hyperparameters · Checkpoints and experiment tracking · Adapter merging and model evaluationBuild: Fine-Tuning Experiment Dashboard
- Day 78Domain dataset preparation · Parameter-efficient LLM tuning · Model card, sharing and inferenceBuild: Fine-Tuned Domain Assistant
Dataset Curation, Alignment, Reasoning and Multimodal AI
Advanced GenAI Engineering
- Day 79Dataset sourcing and provenance · Licensing and consent · Privacy, PII and domain restrictionsBuild: Dataset Governance Checklist
- Day 80Deduplication and language detection · Quality scoring and filtering · Toxicity, corruption and contamination checksBuild: High-Quality Dataset Curator
- Day 81Instruction datasets · Preference and reasoning datasets · Templates, splits and benchmark contaminationBuild: Instruction and Preference Dataset Builder
- Day 82Supervised alignment · Reward models and RLHF · PPO and DPO foundationsBuild: Preference Alignment Experiment
- Day 83Reasoning-task decomposition · Self-consistency and verifier models · Tool-assisted and outcome-supervised reasoningBuild: Verifier-Guided Reasoning System
- Day 84Vision-language models · Multimodal embeddings · Diffusion and cross-modal application workflowsBuild: Multimodal Study Assistant
Evaluation, Responsible AI and Production Deployment
LLMOps and Production Engineering
- Day 85Task-specific and semantic metrics · Golden datasets and evaluation rubrics · Human and model-assisted evaluationBuild: LLM Evaluation Harness
- Day 86Retrieval precision and recall · Faithfulness and answer relevance · Agent tool-use and task-completion metricsBuild: RAG and Agent Benchmark Suite
- Day 87Bias and harmful outputs · Prompt injection and data leakage · Red teaming, privacy and responsible AIBuild: GenAI Security Assessment
- Day 88FastAPI prediction endpoints · Gradio or Streamlit interfaces · Streaming, sessions and user feedbackBuild: Interactive GenAI Web Application
- Day 89Dockerfiles and containers · Secrets and environment configuration · Cloud model and application deploymentBuild: Containerised GenAI Service
- Day 90Prompt, data and model versioning · Logging, tracing and monitoring · Quantisation, batching, caching, CI/CD and rollbackBuild: Monitored GenAI Delivery Pipeline
Sprint 1 Technical Capstone
End-to-End Generative AI Product
- Day 91User problem and stakeholder definition · Functional and non-functional requirements · Model, retrieval and deployment architectureBuild: Capstone Architecture Blueprint
- Day 92Data acquisition and governance · Document ingestion and chunking · Embeddings, vector index and evaluation setBuild: Capstone Knowledge Foundation
- Day 93Baseline prompts and model inference · RAG and source grounding · Conversation and structured-output workflowsBuild: Capstone RAG Assistant
- Day 94Fine-tuned adapter or specialised model · Tool calling and reasoning workflow · Multimodal or agentic enhancementBuild: Capstone Intelligence Layer
- Day 95Evaluation and red teaming · API, UI and Docker integration · Cloud deployment and monitoringBuild: Production Capstone Release
- Day 96Model and system documentation · Live demonstration and architecture defence · Technical viva and improvement roadmapBuild: KnowledgeForge GenAI Platform