AI First Product Engineer Wiki

Theory and Foundation Layer

• math fundamentals
• CS and programming fundamentals
• AI fundamentals
• LLM fundamentals

LLM Train

• MLOps

• pre-training
• post-training

• LLM knowledge distillation

LLM Inference

• GPU resource management
• API / SDK encapsulation
• rate limiting
• error handling
• logging
• monitoring
• alerts
• notifications

LLM Fine-tune

• prefix fine-tuning, prompt tuning, variants
• SFT
• RLHF / RLAIF / DPO variants
• LoRA and QLoRA variants

LLM RAG

basic patterns

• dense vector-based RAG
• sparse vector-based RAG
• graph-based RAG

SOP

• ingest documents, chunking and embedding (structured data) with strategies
• recall with hybrid search
• format, references and citations
• re-rank, query-rewriting, multi-hop, graph or table augmentation
• composable and modular RAG system architecture
• domain-specific retrieval pipelines; continuous ingestion
• quality metrics, evals and quality dashboards

LLM Prompting Engineering

Context Engineering

• prompting engineering BP for human
  • classic patterns: one / few shots, chain-of-thought, self-consistency, reAct etc.
• context summary and chunking
• prompt compression (information compression)
• pick and compose right LLMs for the task
  • model family selection
    • open-source LLMs family
    • commercial LLMs family
  • latency, cost, throughput, quality, etc.
• LLM parameters (tokens, top-p, temperature, etc.)
• prompt management (version, testing, validation, safety, etc.)
• AI driven prompting optimization (prompting refine by AI and auto.)
  • DSPy, textGuard, promptWizard, GRAD-SUM, ell, StarGo ...

LLM Agentic Systems

• basic patterns:
  • CoT
  • ReAct
  • passive goal creator
  • proactive goal creator
  • prompt / response optimizer
  • RAG
  • single / multi path plan generator
  • self-reflection and refinement
  • cross-reflection
  • human reflection
  • voting / role / debate based cooperation
  • tool / agent registry
  • tool execution sandbox
  • agent evaluator
  • multi-modal guardrails
  • ...
• basic principles for agent build
• human-in-loop
• memory
  • short-term memory v.s. long-term memory
    • graph-based v.s. tree-based
    • vector store v.s. graph db v.s. relational db
    • file systems
• context-sizing control
• tool-call and skills management
  • code execution
  • html / web-page (stack) generation
  • browser-use
  • vm use
  • web search
  • ...
• multi-step workflow
  • traditional multi-step workflow
  • claude skills (fixed patterns as sub-agent in similarity)
• agentic-flow prompting
  • ReAct agent
  • reflection x planning x action
  • RPA loop: perception x reasoning x action loop
  • ...
• user-interface customization
• knowledge and RAG enhancements
• continuous learning loop (telemetry → evals → prompt/knowledge updates)
• metrics (cost, latency, throughput, prompting logs, tool-call logs, etc.)
• agent Hallucination prevention and mitigation
• safety, security, compliance, governance
  • content filters, PII redaction, secure key management
  • prompt injection defenses, retrieval hygiene, tool permissioning
  • policy layers (allow/deny lists), sensitive actions with human approval
  • compliance processes (data retention, audit trails), red-team exercises
• performance and cost optimization
  • token budgeting, caching, short prompts
  • reranking before generation, response compression, approximate search tuning
  • distillation/routing to small models; speculative decoding
  • SLAs with adaptive quality tiers, cost/perf dashboards
• agentic mesh
  • memory share and management among agents
  • centralized control v.s. de-centralized and self-organized
  • hierarchical v.s. flatten
  • serial v.s. parallel
  • supervisor v.s. none-supervisor
  • communication protocol
    • end-to-end
    • broadcast
    • shared-memory-channels
  • state-based v.s. memory-based
    • short-term memory v.s. long-term memory
    • graph-based v.s. tree-based
    • vector store v.s. graph db v.s. relational db
  • tool invocation protocol -> MCP (model context protocol)
  • human interfere in agentic loop
    • human as supervisor
    • human as part of the loop
    • human as meta-agent

LLM Product Engineering

Classic Protocols

• MCP (Model Context Protocol)
• A2A (Agentic to Agentic Protocol) with ADK
• Ag-UI (Agentic UI Protocol)
• Agent to Editor (Client) Protocol

Frameworks

• ai-sdk (node / javascript)
• LangChain (python)
• LangGraph (python)

Platforms

Model Services Vendors:

• Open Router
• Claude / Gemini / Grok / OpenAI / DeepSeek / ...

LLM Orchestration Platforms:

• OpenAI Agent Builder
• Dify / Coze
• n8n

Observation

• LangSmith

Test and Evaluation

• Langfuse
• PromptFoo

LLM Deep Scenarios

AI First product systems

VibeCoding

• basic principles and manifesto

OpenSource research:

• Gemini CLI
• Cursor

Arno's BP for VibeCoding

Manus - General Agentic System

patterns:

• monolithic
• pipeline sub-systems
• multi-agent sub-systems (MoA)
• hybrid mixed

info resources:

• domain-specific / public information retrieval

context:

• memory management
• context management / compress and optimize

plan strategies

• static workflow
• intent to plan
• unified intent planning

OpenSource research:

• OpenManus

DeepResearch

• OpenResearch

NoteBook

• Google Notebook ML

MultiModal

• Gen Image
• Gen Video
• Gen Audio
• Gen 3D objects

Reference

• LLM dev stack guide