GigLift: Hinglish job requests in, clear briefs out (Gemma + Django)

Описание и применение

GigLift преобразует неструктурированные сообщения от владельцев магазинов в четкие брифы, уточняя бюджет, сроки и требования, что упрощает работу студентов-фрилансеров.

Задача

Shop owners in India send informal WhatsApp messages for small jobs, leaving out essential details like budget, deadline, and deliverables, causing confusion and costly revisions for students who act as freelancers.

Результат

GigLift successfully transforms informal job requests into structured briefs, improving clarity for both clients and students. Rule-based extraction ensures accurate handling of budgets and deadlines, while the AI generates deliverables and questions. The prototype runs on Render’s free tier and has been evaluated with held-out briefs, achieving high accuracy for budget and deadline extraction.

Рабочий процесс

• 1. Client pastes WhatsApp-style request into GigLift.
• 2. Core AI module (Gemma) parses the text; rule-based extractor reads any stated budget or deadline and prevents the model from altering them.
• 3. If budget or deadline missing, the model provides a rough guess and prompts the client for clarification.
• 4. The system generates deliverables and open questions based on the request.
• 5. Client reviews and edits the brief; upon publishing, it becomes visible to students.
• 6. Students view ranked gigs, submit proposals, and the client hires one.
• 7. Phone numbers and payment details are revealed only after the hire.

Доказательства

GigLift turns it into a proper brief: what has to be delivered, which skills it needs, a budget, a deadline, and the questions the owner forgot to answer.

Источник

If the key is missing or the call fails, rule-based fallbacks take over and the page says so.

Источник

I wrote 30 briefs by hand (English, Hinglish, Devanagari) and built the rules on them. Then I froze extract.py, whose hash is in eval/FROZEN.txt, wrote 32 new briefs and scored them once.

Источник

The five misses were paanch hazar (an amount in words), 1800 mein ho jaye, due 25 Oct, and within 48 hours. A miss means the app asks the client.

Источник