The handbook provides participants with the key technical resources, sponsor tools, project ideas, and best practices needed to successfully build and demo projects aligned with the hackathon’s theme.

Hackathon Handbook — Real‑Time Search & Automation

Welcome — build fast, demo runnable code, measure impact. This handbook gives focused references, stacks, project ideas and a demo checklist so you can deliver an end‑to‑end realtime search + automation prototype within the hackathon window.

Quick links

• Event page & RSVP: https://paris.aitinkerers.org/p/ai-hackathon-paris-2025-real-time-search-automation-hybrid

What to deliver (must-haves)

1. Runnable demo (docker-compose or simple start script) and README with run steps, ideally a working website or app.
2. Architecture diagram + short data flow description.
3. Short demo script (2–4 minutes) and verification steps for judges.
4. Link to source repo (public or private with access credentials).

Judge rubric (make it easy for judges to score)

• Running Code (30%)
• Innovation & Creativity (25%)
• Real‑World Impact (20%)
• Theme Alignment (15%)
• Bonus (10%): Working app/ website

Suggestion: Include a short README section called JUDGE.md that maps your demo to these categories.

Project ideas (fast to prototype)

• Live FAQ operator: stream incoming customer messages, retrieve related docs, auto-suggest replies and optionally send or flag for review.
• Monitoring-to-action: ingest application logs, detect anomaly via embeddings + vector search, have an agent triage and open a ticket or call a webhook.
• Meetings summarizer + task extractor: real‑time transcript ingestion, streaming vector index, generate action items and create tasks in a project board.
• Competitive intelligence pipeline: monitor public RSS/feeds/tweets, index, notify or auto-compose a short brief.

Pick one clear automation endpoint — judges want to see the action completed.

Data, privacy & practical tips

• Use synthetic or public data when in doubt about privacy. If you must use private data, redact PII.
• Keep API keys out of repos (use environment variables or secrets). Provide a test key or instructions for judges.
• Instrument costs: LLM calls add up — use smaller models during development and scale up only for final demo.

Demo checklist (run this before submission)

• README + JUDGE.md present and clear
• Docker or start script works on a fresh machine
• Architecture diagram included
• Demo script and expected outputs (2–4 minutes)
• Repo link and access provided to organizers
• Short video (optional) showing the demo running locally (helps if live demo fails)

Quick starter template (commands)

1) Clone template repo and start local stack:

git clone <your-repo>  
cd demo  
docker-compose up --build  
# or  
./start.sh  # document equivalent steps in README  

2) Run a sample ingest + query script:

python scripts/ingest_sample.py --source samples/news.json  
python scripts/query_demo.py --q "Summarize the latest alert"  

Add a smoke_test.sh that runs a single ingest and query and exits with non‑zero code on failure — judges will appreciate this.

Where to get help during the event

• Onsite: look for organizers.
• Online: the event Message Board / Mentors office hours links (check your RSVP confirmation). Provide a direct GitHub link when asking questions.

Final tips

• Prioritize a small scope: one clear automation loop that is reliable and measurable.
• Skip polishing UI — judges focus on reproducible backend and measurable outcomes.

Good luck — ship something you can demo in 4 minutes and show how it scales. See you at the demo stage!