AI Engineering Masterclass

You'll Build: Website Summarizer with LLM Configuration Playground

Production Goals:

• Implement custom BPE tokenizer, logits adjustment, and major decoding methods.

LLM Techniques to Master:

• Perform Common Crawl & Heuristic filtering.
• Build a BPE tokenizer to map text to tokens.
• Adjust logits via logits bias, temperature, and repetition penalty.
• Interactively apply stochastic/deterministic decoding methods.
• Deploy the inference via an API as a microservice.

You'll Build: Notion Smart Clipper (Chrome Extension)

Production Goals:

• Architect low-latency, multi-source retrieval pipelines.

LLM Techniques to Master:

• Late Interaction (ColBERT) for token-level retrieval.
• Semantic caching for realtime responses.
• Async data ingestion for real-time workspace syncing.
• OpenLLMetry for retrieval pipeline tracing.

You'll Build: A Science-Backed Personal Weight Trainer

Production Goals:

• Eliminate hallucinations in domain-specific tasks.

LLM Techniques to Master:

• Hybrid Retrieval Fusion (BM25 + Dense Vectors)
• Cross-encoder reranking to filter for noisy data
• Structured extraction with Pydantic for JSON outputs
• Self-correction loops via Corrective RAG (CRAG)
• CI/CD for automated accuracy regression tests

You'll Build: Your Digital Clone (Chatbot)

Production Goals:

• Encoding unique personality traits into model weights.

• Squeeze a massive model as on-device AI without needing a $30,000 GPU.

LLM Techniques to Master:

• Model distillation with GPT-4o as a teacher for Llama-3-8B
• Direct Preference Optimization (DPO) for style transfer
• Chat template engineering using Alpaca vs ShareGPT vs Llama3
• Model quantization using GGUF vs AWQ for 4-bit edge deployment
• vLLM serving for optimizing throughput for concurrent persona chats

You'll Build: A Hotel Booking Assistant

Production Goals:

• Implementing persistent agency that survives server crashes.

• Managing high-stakes tool calling with human-in-the-loop.

LLM Techniques to Master:

• QLoRA fine-tuning for high-precision tool (function) calling.
• Parallel tool execution for complex multi-step tasks.
• Adapter-based training to eliminate JSON formatting hallucinations.
• Interrupt-driven workflows (Human-in-the-loop).
• Fault-tolerant retry logic for API failures.

You'll Build: Vibe Coding Assistant

Production Goals:

• Maximize logical depth to solve complex problems with Tree-of-Thoughts.

LLM Techniques to Master:

• Tree-of-Thoughts (ToT) for exploring multiple reasoning branches
• Self-reflection & critique
• Sequential workflows from Planning -> Execution -> Verification
• Adversarial multi-agent debate to find consensus through conflict
• Inference-time compute using DeepSeek-R1 style reasoning

You'll Build: A Smart Fridge App

Production Goals:

• Achieve accurate object detection and inventory matching in unconstrained physical environments.

LLM Techniques to Master:

• Fine-tuning the multi-modal projector for object alignment with Vision-LoRA:
• Multi-modal embeddings for searching by image similarity
• Optimizing multi-modal inference latency with TensorRT-LLM
• Measuring the accuracy of image-text alignment with CLIP score
• Video latent caching: Exploring frame-by-frame inventory tracking

You'll Build: An Enterprise LLM Guardrails & Cost-Per-Query Dashboard

Production Goals:

• Transit from prototypes to enterprise systems by integrating real-time guardrails and unit-economic tracking.

LLM Techniques to Master:

• LLM-as-a-Judge for automating regression testing and output validation.
• Automated PII redaction and data masking.
• Token ROI modeling for the cost-to-value analysis
• Prompt versioning with Git-based prompt management
• Red-teaming simulation over prompt injection and jailbreak attacks