Intelligent Candidate Discovery — Model Artifacts
Pre-computed retrieval artifacts for the Redrob AI — India Runs Data & AI Challenge (Track 1) candidate discovery and ranking system.
This repository contains the FAISS search index and associated metadata required by the deployed application for semantic candidate retrieval.
Artifacts
| File | Purpose |
|---|---|
| aiss.index | 768-dimensional FAISS IndexFlatIP semantic search index |
| candidate_lookup.pkl | Candidate ID → candidate profile metadata lookup |
| embedding_metadata.pkl | Metadata associated with the generated candidate embeddings |
The embeddings were generated using BAAI/bge-base-en-v1.5.
Usage
The production application downloads these artifacts automatically when they are not available locally.
Live application:
https://huggingface.co/spaces/ParminderzHuggingFace/redrob-ai-candidate-ranking
The candidate profile database (candidates.jsonl) is maintained separately in the Hugging Face Dataset repository:
https://huggingface.co/datasets/ParminderzHuggingFace/india-runs-candidates
Architecture
The deployment separates source code, candidate data, and pre-computed retrieval artifacts:
- GitHub (ParminderSinghGithub/India-Runs) → application source code
- Hugging Face Dataset (ParminderzHuggingFace/india-runs-candidates) → candidates.jsonl
- Hugging Face Model (ParminderzHuggingFace/intelligent-candidate-discovery-models) → FAISS index and embedding metadata
- Hugging Face Space (ParminderzHuggingFace/redrob-ai-candidate-ranking) → live Streamlit application
The artifacts in this repository are pre-computed offline and are not regenerated during application startup.