# Minimun Developer
## Definition
- 專案完成所需技能, 學完後可以完成組織交付的任務
## DoD
- 能開發新的feature(function/method) (Check Point)
## Flow
- [1]能在local起專案
- [2]能寫出符合團隊規範的code
- [3]能操作符合團隊的git流程

## Must Read
- [Google Python Style Guide](https://google.github.io/styleguide/pyguide.html)

## [1] IDE
- Pycharm
- DataSpell

## [1] fileformat
- yaml
- json
- .env

## [1] Environment Management
- Poetry
- Pyenv (Python版本管理)
- Docker 基礎指令
- Docker compose 基礎指令

## [1] Going Cloud
- Set up Going Cloud Project Template
- Can run a chatbot and use it

## [2] OOP
- Classes
- Inheritance
- Methods

## [2] Package / Library / Framework
- Gradio
- boto3
- FastAPI
- Pydanic

## [2] Development principle (懂基礎概念)
- Controller
- Service
- Entity
- Repository
- Gateway

## [2] Design principle
- SOLID
- DRY - Don’t Repeat Yourself
- KISS - Keep It Simple & Stupid
- YAGNI - You Ain’t Gonna Need It

## [2] Others
- Type hints

## [3] Git
- Basic(Add, Commit, Branch, Push, Pull, Merge, Rebase, Stash)
- Branch Strategy (Trunk Based Development)
- Conventional Commit
- 確定pre-commit有執行

## [DoD] Going Cloud
- New Feature to Going Cloud Example Project

# Minimun IR-based LLM Application(GC-IRLLM) （IR＝Information retrival）
## Definition
- 專案完成所需技能, 學完後可以完成組織交付的任務, 尤其實在專案基礎上微調

## DoD
- New Toy Project of GC-IRLLM

## Flow
- [1] 知道IR-based LLM Application概念
- [2] 能在local起專案
- [3] 能call 專案內API 餵資料進去
- [4] 能調整prompt
- [5] 有基礎QA能力

## [1] Going Cloud
- Set up GC-IRLLM Template

## [2] LLM setting (MUST)
- Temperature
- Top P
- Other Hyper parameters

## [2] LLM concept (MUST)
- Types of LLMs
- Pros and Cons of LLMs

## [2] Application Concept
- What's RAG
- Standard RAG flow

## [2] Package / Library / Framework (MUST)
- LangChain
- Azure OpenAI API
- AWS BedRock API
-
## [3] Going Cloud
- GC-IRLLM Data API
- Basic ETL from Raw Data

## [3] Data Chunk Strategy
- Whole
- Character Text Split

## [4] Prompting Techniques
- Role Prompting
- Tasks Prompting (QA/Summary/Generation)

## [4] Functionality
- Question Answering

## Debug
- [5] 知道如何操作GC-IRLLM, 查看檢索到的資料
- [5] 知道基礎實驗記錄方法(excel, o11y service)

## Demo
- [DoD] Demo new toy IRLLM

---

# Basic Developer(團隊最小公約數 (junior))

## Development principle
- Clean Architecture

## Environment Management
- Poetry
- Pyenv (Python版本管理)
- Docker 基礎指令
- Docker compose 基礎指令

## Package / Library / Framework
- Huggingface Transformer
- Pandas
- Numpy
- Scikit-learn
- Pytorch or Tensorflow

## Others
- Error handling

## Linting/Formating Settings
- black
- isort
- Flake8
- pylint
- EditorConfig

## RDBMS tools
- SQLalchemy
- SQL syntax (這個要會)

## API Document
- Swagger (in FastAPI)(不會也可以開發)

## API
- RestfulAPI

# Advanced Developer

## Git
- Git hook

## Profling and Performance Debug
- pyroscope
- Memray
- Viztracer
- profyle

## Optimization skill
- asyncio (async/await)
- Thread
- Multi-process

## Testing
- Integration Testing (high level to API Testing)
- Test Driven Development
- Domain Driven Design
- UI Testing

## Testing
- Unit testing

## Test Tools
- Pytest

# Basic Machine Learning (MLE 的基本素養)
## Task
- Classification
- Clustering
- Regression

## Training Strategy
- Supervise Learning
- Un-Supervise Learning
- Semi-Supervise Learning

## Knowledge
- Machine Learning (基本概念)
- Deep Learning (基本概念)

## Neural Network
- ANN
- VAE
- Transformer (Encoder, Decoder)

## Domain knowledge
- GenAI

## Design and Development Principles
- Golden dataset Desgin
- Experiment Design

# Advanced Machine Learning (MLE 的基本素養)
## Knowledge
- Reinforcement Learning
- Online learning

## deploy skill
- Distilling
- Quantization

## Domain knowledge
- NLP
- RecSys

# Basic IR-based Application

## Definition
- 團隊最小公約數 (junior)
- 能解釋啥是好, 啥是壞

## Application Concept
- Relevant search
- Vector Similarity Function

## Retrieve Sources (會打基本的 API)
- Vector DB (OpenSearch/ElasticSearch)
- Document DB (OpenSearch/ElasticSearch)
- Files (PDF, ppt ...etc)

## Tools
- opensearch-py/elasticsearhc-py

## Retrieve Strategy
- Single
- Fusion

## Generation Strategy
- text
- streaming
- prefix, postfix

## Data Chunk Strategy
- Whole
- Character Text Split

## Functionality
- Summarization
- Question Answering

# Advance IR-based Application

## ElasticSearh / OpenSearch (No)
- Vector Search
- Text Search
- Index
- Term Query
- String Query
- Analyzer
- Keywords

## LLM usage concept
- Agents
- Tools
- Characters

## IR Concept
- RAG
- REALM
- RETRO

## NLP Concept
- NER
- NLU
- QA

## Package / Library / Framework
- SemeticKernel (??? 為什麼要會這個)

## Retrieve Sources
- NoSQL (DynamoDB, levelDB, Redis)
- SQL (DuckDB)
- Graph DB (neo4j/arrangodb)

## Data Chunk Strategy
- Document QA generation
- Parent-Child
- Knowledge Graph

## Prompting Techniques
- Few Shot Learning Prompting
- Chain of Thought
- Zero Shot Learning Prompting
- Chain of Validation
- Chain of Note
- Tree of Thoughts
- Rephrase and Respond
- Personalized (NLU, status) (組合型的 prompt 來達到個人化的效果)

## Retrieve Strategy
- [Rerank] Reciprocal Rank Fusion
- [Rerank] Cohere

## Generation Strategy
- chunk streaming (研究中)

## Functionality
- Summarization
- Multiple rounds of Dialog
- Content Generation

## Evaluation
- Context Relevancy(Recall/Precision/F1)
- Generated Answer Relevancy

## Evaluation Tools
- Ragas
- promptfoo

## LLMOps
- Prompt Versioning
- Index Versioning
- Data Versioning

## LLMOps Tools
- [Experiment]MLflow
- pezzo/BentoML
- Grafana + Prometheus
- LIDA
- llmonitor (langfuse?)

## Networks
- CORS
- HTTPS
- OAuth
- JWT

## Others
- gunicorn/unicorn