---
title: MLOps RAG Agent
emoji: 🤖
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 6.18.0
app_file: app.py
pinned: false
license: mit
python_version: "3.10"
short_description: Level 3 Agentic RAG for MLOps/DevOps Q&A
---

# 🤖 MLOps / DevOps Agentic RAG Agent

A **Level 3 Agentic RAG** pipeline for MLOps and DevOps Q&A, featuring query rewriting, iterative retrieval, self-reflection, source citation, and fallback handling.

## Architecture

```
User Query
    │
    ▼
┌─────────────────────────────────────────────────────────┐
│               Level 3 Agentic RAG Pipeline               │
│                                                         │
│  1. Query Rewriting (Flan-T5-base)                      │
│     └─ Reformulates query for better retrieval          │
│                                                         │
│  2. Semantic Retrieval                                  │
│     ├─ Embeddings: all-MiniLM-L6-v2 (384-dim)          │
│     ├─ Vector Store: ChromaDB (cosine similarity)       │
│     └─ Indexing: LlamaIndex (SentenceSplitter)          │
│                                                         │
│  3. Relevance Filtering (Self-Reflection)               │
│     └─ Flan-T5 checks if each chunk is relevant        │
│                                                         │
│  4. Answer Generation (Flan-T5-base)                   │
│     └─ Context-grounded answer synthesis               │
│                                                         │
│  5. Answer Self-Reflection (Flan-T5-base)              │
│     └─ Judges completeness → triggers re-retrieval     │
│                                                         │
│  6. Iterative Retrieval (up to 2 iterations)           │
│     └─ Follow-up queries if answer is inadequate       │
│                                                         │
│  7. Fallback Handling                                   │
│     └─ General answer when no relevant docs found      │
│                                                         │
│  8. Source Citation                                     │
│     └─ Source doc + cosine similarity per chunk        │
└─────────────────────────────────────────────────────────┘
    │
    ▼
Gradio UI (Answer + Query Info + Reflection + Citations)
```

## Tech Stack

| Component | Technology |
|-----------|------------|
| Orchestration | LlamaIndex Core |
| Vector Store | ChromaDB (persistent) |
| Embeddings | sentence-transformers/all-MiniLM-L6-v2 |
| Generator | google/flan-t5-base |
| UI | Gradio |

## Knowledge Base

The curated knowledge base covers five MLOps/DevOps domains:

- **AWS SageMaker** — Training jobs, HPO, endpoints, Pipelines, Feature Store, Model Monitor, Clarify
- **Kubernetes** — Pods, Deployments, HPA/KEDA, GPU management, Kubeflow, KServe, Helm
- **CI/CD** — GitHub Actions, Jenkins, ArgoCD/GitOps, DVC, MLflow model registry, testing strategies
- **Terraform** — HCL syntax, remote state, modules, EKS/SageMaker provisioning, Terragrunt
- **Model Monitoring** — Data drift, concept drift, Evidently AI, Alibi Detect, Prometheus/Grafana, Great Expectations

## Example Questions

- How do I configure auto-scaling for a SageMaker inference endpoint?
- What is the difference between data drift and concept drift?
- How do I use Terraform to provision an EKS cluster with GPU nodes?
- Explain how to set up a CI/CD pipeline for model retraining with GitHub Actions.
- How do I deploy a model with canary rollout on Kubernetes using KServe?
- What metrics should I monitor for ML models in production?

## Running Locally

```bash
git clone https://huggingface.co/spaces/atulkrs/mlops-rag-agent
cd mlops-rag-agent
pip install -r requirements.txt
python app.py
```