mishface123/finance-intelligence-ascended

Team Ascended - AutoScientist Challenge (HackIndia)

Official workspace repository for Team Ascended competing in the AutoScientist Challenge × HackIndia 2026. This repository contains the automated execution pipeline, localized configurations, and interactive Gradio interface deployed to Hugging Face Spaces.

🪐 Project Overview

Our project focuses on Llama-4-Scout-Finance-Ascended, a specialized financial intelligence engine optimized for zero-shot quantitative reasoning, context extraction, and regulatory compliance tailored specifically to the Indian economic ecosystem.

Key Performance Metrics:

• —Quality Improvement: +51.7% relative quality improvement over the foundational baseline.
• —Domain Win Rate: 70% win rate on complex corporate and financial QA datasets.
• —Evaluation Grade: Upgraded from a baseline 'C' to a production-ready 'A'.

🔗 Project Repositories & Hubs

• —Hugging Face Model Card: mishface123/llama-4-scout-finance-ascended
• —Hugging Face Dataset: mishface123/adaption-econ-finance-qa-pairs
• —Live Interactive Demo (HF Space): mishface123/finance-intelligence-ascended
• —Kaggle Model Hub: llama-4-scout-finance-ascended on Kaggle

🧬 Tracking & Execution Architecture

1. Data Ingestion & Mutation Pipeline

• —Trained Model ID: adaption_llama_4_scout_17b_16_econ_finance_qa_pairs_b35271f3
• —Evolved Dataset ID: 14ef7f3a-4294-418a-bfe6-7702472c893d
• —Adaption API: Programmatic execution pipeline located at `scripts/run_pipeline.py`.
• —Environment Configuration: API keys are dynamically loaded at runtime from the environment variable:

  export ADAPTION_API_KEY="pt_live_..."

2. Standalone Web Dashboard

The interactive user interface is built on Gradio and hosted under the standard Gradio SDK runtime environment. The UI permits users to:

• —Submit quantitative financial reasoning queries.
• —Query the Adaption Labs inference endpoint.
• —View structured analytical responses from Llama-4-Scout-Finance-Ascended.

Refer to `demo/app.py` and `demo/requirements.txt` for implementation details.

🛠️ Workspace Directory Layout

.
├── config/
│   └── hyperparameters.json       # AutoScientist training configuration payload
├── demo/
│   ├── app.py                    # Gradio dashboard UI script
│   └── requirements.txt          # Space runtime dependencies
├── scripts/
│   ├── data_streamer.py          # Python memory-efficient dataset streaming script
│   └── run_pipeline.py           # AutoScientist data ingestion/mutation execution pipeline
├── LICENSE
└── README.md                     # Main project documentation

🧑‍💻 How to Run the Pipeline

Ensure your environment variables are configured, then execute:

python scripts/run_pipeline.py

To run the interactive Gradio demo locally:

pip install -r demo/requirements.txt
python demo/app.py