Forward Deployed Engineering
19 topics · Stage 6
Complete
- 01Introduction to Forward Deployed Engineering and GenAI FoundationsWhat a Forward Deployed Engineer actually does, why "build a chatbot" is not a requirement, and the road from rule-based AI to machine learning, neural networks, attention and GPT.24 min
- 02How Large Language Models WorkFollow one sentence through an LLM. Tokens, token ids, embeddings, position, attention with query, key and value, Transformer layers, logits, softmax, and how decoding picks the next token.13 min
- 03Talking to an LLM from Your ApplicationAn LLM is a capability, a GenAI app is software built around it. Call a model from Postman, read the response and token usage, then build a ticket summariser in Spring Boot with Spring AI, Node.js and Python.14 min
- 04Building a Chatbot: Context, Memory, Streaming and HallucinationsA model call is not a conversation. Roles, system prompts, app memory, the context window, streaming and hallucination, with a support chatbot built in Spring AI, Node.js and Python.23 min
- 05Introduction to AI Agents: Tools, Tool Calling and the Agent LoopAn LLM writes text but cannot act. Tools give it hands. Tool definitions, the tool calling loop, information vs action tools, and a sandboxed website builder agent in Spring AI, Node.js and Python.27 min
- 06Streaming AI Responses: From Tokens to a Live ChatWhy a reply that streams feels faster even when it is not. Generation vs delivery, time to first token, chunked HTTP, SSE vs WebSockets, Flux in Spring AI, saving history after the stream, and reading it in the browser.29 min
- 07Vector Embeddings from First PrinciplesHow a computer can hold meaning as numbers. Vectors, dimensions, distance, dot product and cosine similarity, why hand made features fail, and how training turns context into embeddings.33 min
- 08Semantic Movie Search with EmbeddingsBuild a small Netflix style search. Embedding model vs chat model, why sending the whole catalogue to an LLM fails, precomputed vectors, top K by cosine, and similar movies, in Spring AI, Node.js and Python.27 min
- 09Vector Search: Brute Force KNN and the Indexing ProblemWhat breaks when the movie search grows from 30 to millions of vectors. Exact vs similarity lookup, brute force top K, the real cost in compute and memory, and why B-trees and k-d trees stop helping in high dimensions.31 min
- 10Vector Databases and ANN: IVF, HNSW and Product QuantizationHow vector databases search millions of vectors fast by not looking at most of them. Approximate search and recall@K, IVF clusters, HNSW graphs, quantization and PQ, metadata filters, and pgvector.38 min
- 11Retrieval Augmented Generation (RAG): Chunking, Retrieval and CitationsA support bot that answers from company policy documents. Chunking, embedding, a pgvector store, retrieval, prompt stuffing with citations, and the ways RAG fails in production.37 min
- 12Advanced RAG: Chunking, Hybrid Search and RerankingWhy basic RAG misses rules and exact product codes, and how to fix retrieval with structural chunking, top K, thresholds, metadata filters, BM25 hybrid search, RRF, reranking and context engineering.40 min
- 13Structured Output: Typed JSON From an LLMTurn one sentence into a typed MeetingDetails object your code can trust. Schemas, strict mode, the checks a schema cannot do, retries with feedback, in Spring AI, Node.js and Python.34 min
- 14Model Context Protocol (MCP): Hosts, Clients and ServersWhy every AI app rewriting the same tools does not scale, and how MCP fixes it. Hosts, clients and servers, tools/list and tools/call, and a task server with an LLM client in Spring AI, Node.js and Python.32 min
- 15MCP Beyond Tools: Resources and PromptsAn MCP server offers more than tools. Resources give the app context, prompts give the user ready workflows. Who controls each, how list and read work, and a task server and client in three stacks.36 min
- 16Multimodal Vision Chat: Sending Images to an LLMHow a model reads an image (patches, a vision encoder, an adapter), what image tokens cost, size limits, upload safety, and a vision chat with a browser page in Spring AI, Node.js and Python.40 min
- 17GenAI Evals: How to Know Your AI App WorksWhy assertEquals breaks on LLM output. Eval and golden datasets, deterministic checks, five quality metrics, LLM-as-a-judge with rubrics, human review, RAG evals, and offline vs online evals.28 min
- 18Testing an AI Agent: Deterministic Evals and LLM-as-a-JudgeAn eval runner for a support agent with RAG and tools: a trace, a golden dataset, four checks in code, a judge with strict scores and a pass threshold, in Spring AI, Node.js and Python.26 min
- 19GenAI Security and Guardrails: Prompt Injection, Jailbreaks and Safe Tool UsePrompt injection, jailbreaks and why a system prompt is not a lock. Five rules for safe tools, layered guardrails, and a guarded support agent in Spring AI, Node.js and Python.33 min