ScholarScout

Inspiration

Researchers — especially students and early-career academics — lose hours hunting for where to submit their work. Calls for papers, conference deadlines, journals, and grants are scattered across dozens of sites, and it's genuinely hard to tell a reputable, indexed venue from a predatory pay-to-publish one. We wanted an agent that does this scouting automatically and, crucially, gives current, trustworthy answers.

What it does

You type your research field. ScholarScout:

1. Uses NVIDIA Nemotron to reason about your field and generate focused search queries.
2. Uses Tavily to search the live web for current CFPs, deadlines, journals, and grants.
3. Uses Nemotron again to judge each venue's reputability (flagging predatory ones) and write a skimmable, grouped digest — Urgent, Open now, Upcoming deadlines, Best early targets — each with deadline, days-left, cost (FREE vs APC), a reputability note, and a clickable link, ending with a suggested submission plan. A deadline-window filter (30/45/60 days) lets you focus on what's urgent.

How we built it

• Nebius Token Factory (OpenAI-compatible API) serves nvidia/nemotron-3-super-120b-a12b for all three reasoning steps.
• Tavily (search_depth="advanced") grounds the model in real, cited sources.
• Python engine (scholarscout.py, also a CLI) + Streamlit UI, deployed on Streamlit Community Cloud.
• Clean separation: the agent degrades gracefully to model-only mode if no search key is present.

Challenges we ran into

The biggest insight came from a failure: running Nemotron alone, it confidently produced plausible but outdated deadlines (stuck at its training cutoff). Adding Tavily grounding turned hallucinated guesses into real, current, cited venues — a vivid demonstration of why agentic search matters for any factual assistant.

Accomplishments that we're proud of

A genuinely useful, trustworthy research tool built on open infrastructure — grounded, reputability-aware, and open-source (MIT).

What we learned

How to compose a reliable retrieval-augmented agent on open models, and how much grounding (Tavily) improves factual reliability over a model-only baseline.

What's next for ScholarScout

Automated recurring digests by email, grant-database integration, personalized ranking by fit, and a reviewer/peer-review-opportunity finder.

Built With

• nebius-token-factory
• nvidia-nemotron
• openai-sdk
• python
• streamlit
• tavily