Status: In progress
Expected Release: 31/10/2025

Will be available by end of october (hopefully) on https://l145.be/ so that you can ask questions to the AI version of me!

• AI-Powered Q&A: Ask questions in plain language and get answers based directly on my skills, projects, experience, and more.
• Custom RAG Pipeline: A custom-built pipeline ensures that the answers are highly relevant and drawn from optimized data chunks.
• Fact-Checked Responses: The AI is instructed to only use the provided context from my portfolio, preventing it from making things up.
• Fast & Efficient: Uses the Groq API for real-time, low-latency responses.
• Portfolio Integration (todo): Direct integration into personal portfolio site https://l145.be/.

1. Ingestion: My portfolio data is manually chunked into focused pieces for optimal embedding quality.
2. Embedding: Each chunk is converted into a vector using the all-MiniLM-L6-v2 model.
3. Storage: These vectors are stored in a ChromaDB vector database.
4. Retrieval: When you ask a question, it's converted into a vector to find the most similar (relevant) data chunks from the database.
5. Generation: Your original question and the retrieved data are sent to the Groq LLM, which generates a coherent, context-aware answer.

• LLM: Groq (using openai/gpt-oss-20b)
• Vector DB: ChromaDB (Local for dev, Chroma Cloud for prod)
• Embedding Model: sentence-transformers/all-MiniLM-L6-v2
• Llama-index (LEGACY): Various library features, scrapped due to bad accuracy and chromadb sync issues, would be better accuracy if there was more data, which is not not applicable to a simple Portfolio RAG

1. Clone the repository.

2. Start the local ChromaDB server:

   # Using PowerShell
   pwsh ./chroma-start.ps1

   or

   # Using bash/terminal
   chroma run --host localhost --port 8000

3. Go through pure_chroma.ipynb if you want to try it yourself.

• Hyper-optimize and manually chunk all portfolio data.
• Build and test the dense RAG pipeline locally.
• Chroma Cloud: Resolve the "permission denied" error to connect to the cloud instance.
• Backend Integration: Build out the backend.
• Frontend Integration + deployment: Design and build the AI chat section for my portfolio at l145.be.
• Deployment (Backend): Deploy backend (somewhere)

• Author: Aryan Shah
• Email: aryan.shah@l145.be
• GitHub: l145dev
• LinkedIn: Aryan Shah