Hey Devs! π
Iβm currently working on an exciting hands-on project that blends the power of LLMOps and DevOps: an AI-powered Anime Recommender System.
Though the project is still a work in progress, I wanted to share what Iβm exploring:
- The architecture and toolchain I'm using π§±
- The technologies Iβm learning along the way π§°
- And why this is such an exciting learning opportunity π‘
I believe in learning in public, so hereβs a peek behind the scenes!
π― Project Goal
The goal is to build a recommender system that
- Uses Groq LLM APIs and HuggingFace embeddings
- Stores vector data in ChromaDB
- Communicates with LLMs using LangChain
- Has an interactive UI built in Streamlit
- Is deployed on GCP VM via Docker + Kubernetes
- And monitored with Grafana Cloud
π οΈ Tech Stack
Category | Tools/Technologies |
---|---|
LLM & Embeddings | Groq API, HuggingFace |
GenAI Framework | LangChain |
Vector Store | ChromaDB |
Frontend/UI | Streamlit |
Containerization | Docker |
Orchestration | Kubernetes (via Minikube) |
Deployment Platform | GCP VM |
Monitoring | Grafana Cloud |
CLI | Kubectl |
Version Control & CI/CD | GitHub |
π Architecture (Work in Progress)
_Hereβs the architectural flow Iβm working
_
Project Setup
- Groq API, HuggingFace API
- Logging, Custom Exception Handling
- Virtual Environment, Project Structure
Core Engine
- Configuration and Data Loading
- Embedding storage using ChromaDB
- Prompt templating
- Recommender Class logic
- Training & Recommendation process
UI and Deployment
- Streamlit App as frontend
- Dockerizing the app
- Deploying on Minikube within a GCP VM
- Monitoring Kubernetes metrics with Grafana Cloud
π What Iβm Learning
β
Integrating Groq and HuggingFace for LLM-based workflows
β
Creating prompt pipelines with LangChain
β
Managing embeddings using ChromaDB
β
Deploying containerized apps using Docker on Kubernetes
β
Running K8s inside GCP VM with Minikube
β
Setting up Grafana Cloud dashboards for cluster observability
β
CI/CD and GitHub-based DevOps flows for MLOps/LLMOps
π Whatβs Next
- Finalizing the embedding + recommender pipeline
- Adding GitHub Actions CI/CD
- Creating APIs for interaction
- Hosting a public demo
- Publishing full documentation and source code
π‘ Why Share This Now?
This isnβt a finished project. Iβm posting while I explore and build β because
- Feedback makes me better
- Others might benefit from this stack
- It keeps me accountable and motivated
π¬ Stay Tuned for Part 2
In the next post, Iβll share
- Source code and repo walkthrough
- API endpoints for recommendations
- LLM prompt tuning strategies
- Live monitoring via Grafana Cloud
Thanks for reading! π
If youβre working on something similar or want to explore LLMOps, feel free to drop your thoughts and ideas in the comments!
Letβs connect π
Top comments (0)