CoolFace
Apppublic

28Shekhar/agentbenchai

sourceHugging Faceupdated 3mo agoView on Hugging Face
0likes
App README

AgentBench AI

AgentBench AI is a deployable course MVP for recommending RAG architecture combinations using a multi-agent workflow.

Important key setup

For this prototype, keys are read from api_keys.py first, then from Hugging Face Secrets if present.

Open api_keys.py and paste:

python
GROQ_API_KEY = "gsk_your_key_here"
LANGSMITH_API_KEY = "lsv2_your_key_here"
LANGCHAIN_API_KEY = ""  # optional; can be same as LangSmith key
LANGCHAIN_TRACING_V2 = "true"
LANGCHAIN_PROJECT = "AgentBench-AI"

Do not make the Space public if you hard-code keys in the repo.

Hugging Face deployment

  1. 1.Create a new Hugging Face Space.
  2. 2.Select SDK: Docker.
  3. 3.Do not select Gradio or Streamlit.
  4. 4.Upload the contents inside this folder, not the parent folder itself.
  5. 5.Commit and wait for build.
  6. 6.Open the app and upload PDF, CSV, TXT, or image files.

What the app does

  • —Ingests PDF, CSV, TXT, and images.
  • —Detects one of 10 domains: Healthcare, Banking, Insurance, Manufacturing, Retail, Telecom, Legal, Education, Pharma, Aerospace.
  • —Runs Embedding, Chunking, Tokenizer, and Indexing agents.
  • —Uses Groq llama-3.3-70b-versatile via chat.completions.create().
  • —Supports HITL approval/rejection and re-run with feedback.
  • —Produces Top-2 RAG architecture combinations.
  • —Adds LangSmith tracing with @traceable wrappers around agent evaluation and Groq calls.