# OPEA 2024 - 2025 Roadmap

## May 2024

### Contribution

- **Components**
  - ASR
  - Data Prep
  - Embedding
  - Guardrails
  - LLM (Gaudi TGI)
  - Rerank
  - Retrieval
  - TTS
  - VectorDB

- **Use Cases/Examples**
  - ChatQnA
  - CodeGen
  - CodeTrans

- **Cloud Native**
  - OneClick OPEA on ChatQnA
  - OneClick OPEA on CodeGen
  - GenAI microservice connector

- **Evaluation & Others**
  - CICD & Validation
  - Eval: E2E (GenAIComps & GenAIExamples), lm-eval-harness, bigcode-eval-harness
  - RAGAS evaluation service

### AI Models

- LLM: llama2 (7b, 13b, 70b), llama3 (8b, 70b), code-llama, Llama guard
- Embedding: BGE-base

### AI Tools Integration

- VectorDB: Chroma
- Framework: Langchain

### Deployment Type

- On Prem,IDC (Xeon, Gaudi)

## June 2024

### Contribution

- **Components**
  - LLM (Xeon vLLM & Ray, Ollama)
  - OVMS
  - prompting
  - user feedback management
  - Mega Component (MI6 RAG service)

- **Use Cases/Examples**
  - DocSum
  - SearchQnA

- **Cloud Native**
  - OneClick OPEA for 2 more examples
  - GMC with switch support (dynamic pipelines)
  - Helm charts/templates for custom yamls (refactoring)

- **Evaluation & Others**
  - CICD & Validation
  - Eval: E2E (GenAIComps & GenAIExamples) Gaudi (2) and CPUs in CICD cluster

### AI Models

- LLM: mistral-7B, mixtral-8x7B
- Embedding: E5-mistral-7b-instruct, all-mpnet-base-v2

### AI Tools Integration

- VectorDB: Pinecone, Redis
- Framework: Llamaindex, Haystack

### Deployment Type

- On Prem,IDC (Xeon, Gaudi)

## July 2024

### Contribution

- **Components**
  - LVM (Gaudi vLLM & Ray)
  - vectordb (svs)
  - Gateway guardrail, Auth Z/N

- **Use Cases/Examples**
  - FAQGen

- **Cloud Native**
  - OpenShift enablement for OPEA
  - OneClick OPEA for 3 more examples
  - Security (Service Mesh, guardrails)

- **Evaluation & Others**
  - CICD & Validation
  - Eval: E2E (GenAIComps & GenAIExamples)

### AI Models

- LLM: Phi, Gemma
- Embedding: all-MiniLM-L6-v2, paraphrase-albert-small-v2

### AI Tools Integration

- VectorDB: PGVector, Qdrant

### Deployment Type

## Aug 2024

### Contribution

- **Components**
  - Documentation
  - Test automation script
  - Telemetry

- **Use Cases/Examples**
  - Documentation
  - Test automation script

- **Cloud Native**
  - Demo K8s resource management
  - Documentation on autoscaler analysis

- **Evaluation & Others**
  - CICD & Validation
  - Eval: E2E (GenAIComps & GenAIExamples)

### AI Models

- Vision: llava
- Mixtral-8x22B

### AI Tools Integration

- VectorDB: Milvus

### Deployment Type

- Public Cloud AWS (Xeon CPU & NV GPU)

## Sep 2024

### Contribution

- **Components**
  - Microservice for Image and Video

- **Use Cases/Examples**
  - Text to Image generation
  - Image to Video generation
  - Playground (composable and configurable)

- **Cloud Native**

- **Evaluation & Others**
  - CICD & Validation
  - Eval: E2E (GenAIComps & GenAIExamples)

### AI Models

- Diffusion model:
  - Stable Diffusion XL
  - Stable Diffusion 3M
  - Stable Video Diffusion

### AI Tools Integration

- VectorDB: Weaviate

### Deployment Type

## Q4 2024

### Contribution

- **Components**
  - Fine-tuning E2E pipeline
  - Knowledge Graph

- **Use Cases/Examples**
  - Fine-tuning (Lora)
  - AI Agent (single Agent with text and Audio as user interface)
  - Closed source LLM
  - GraphRAG

- **Cloud Native**
  - Static tuning on Resource management for deployment

- **Evaluation & Others**
  - CICD & Validation
  - Eval: E2E (GenAIComps & GenAIExamples)

### AI Models

- LLM open: Grok 1
- LLM Close: GPT3.5/4/4o, Claude 3/3.5
- AWS Bedrock endpoint

### AI Tools Integration

- Knowledge graph: Neo4j
- Agent: LangGraph

### Deployment Type

- Public Cloud (Azure, GCP, Oracle, AWS)
- AI PC (Intel)

## Q1 2025

### Contribution

- **Components**
  - more Microservice request from community
  - Confidential Container

- **Use Cases/Examples**
  - AI Agent (Multi Agent)
  - Fine-tuning (Adpative)
  - Long context window (>1M)
  - GenAI Studio

- **Cloud Native**
  - Dynamic tuning on Resource management through K8s

- **Evaluation & Others**
  - CICD & Validation
  - Eval: E2E (GenAIComps & GenAIExamples)

### AI Models

- LLM: SetFit
- More to be defined

### AI Tools Integration

- AutoGen, CrewAI

### Deployment Type

- Public Cloud (tier2 CSP)
- AI PC (others)