Annonce


[Stage] Design and Development of a Conversational Assistant for Social Event Discovery Using Lightweight NLP and LLM APIs

01 Octobre 2025

Catégorie : Postes Stagiaires ;

1. Context

Declic is a rapidly growing social platform designed to help users discover and organize social events and training workshops. With over 100,000 users and thousands of events, the platform aims to enhance user engagement and experience through an intelligent conversational assistant. This assistant will simplify event discovery, streamline user onboarding, and facilitate interactions with the platform.

For more information or to download the app, visit https://declic.net.

This internship is offered in collaboration with Université Paris-Est Créteil (UPEC) and the research laboratory LISSI (Laboratoire Images, Signaux et Systèmes Intelligents).

The proposed project focuses on developing a chatbot assistant that leverages pre-trained language models (LLMs) and lightweight NLP techniques to deliver personalized, real-time, and contextual support—without the need for training large models from scratch.

2. Objectives

The main goal of this internship is to develop a smart, user-friendly chatbot for Declic that can:

  • Interact naturally with users.
  • Recommend events based on user interests and location.
  • Assist users in creating, managing, or finding events.
  • Guide new users through onboarding.
  • Answer frequently asked questions (FAQ).
  • Maintain a lightweight, scalable, and cost-efficient design suitable for real-world deployment.

3. Approach & Methodology

3.1 Intent Recognition

  • Leverage Rasa NLU, spaCy, or DistilBERT.
  • Classify user inputs into categories such as event search, help, create event, or how-to.

3.2 FAQ & Knowledge Search (RAG)

  • Index Declic FAQs, platform tutorials, and event metadata.
  • Use vector search (FAISS or Weaviate) for retrieval-based response generation.

3.3 LLM Integration

  • Integrate with OpenAI GPT-4, Mistral, or Claude for flexible responses.
  • Use LangChain or LlamaIndex for prompt handling and memory management.

3.4 Recommendation Engine Connection

  • Connect the chatbot to a simplified event recommender.
  • Pass user embeddings or filters based on behavior/preferences.

3.5 Optional UI/Integration

  • Connect the chatbot to Declic’s app via FastAPI.
  • Deploy a test version for user evaluation.

4. Technologies & Tools

  • Programming & Backend: Python, FastAPI, pandas, scikit-learn
  • NLP & ML: spaCy, NLTK, Hugging Face Transformers
  • Chatbot Frameworks: Rasa, BotPress, LangChain, Dialogflow
  • LLM APIs: OpenAI, Mistral, Claude
  • Vector Search & Retrieval: FAISS, Weaviate, LlamaIndex

5. Expected Deliverables

  • A working chatbot prototype addressing defined use cases.
  • A modular backend for intent recognition and dialogue management.
  • Integration with real or simulated Declic data.
  • Comprehensive technical documentation and results presentation.
  • Optional: Small-scale user testing or demo with feedback collection.

6. Learning Outcomes

  • Design of modern NLP pipelines.
  • Prompt engineering and LLM-based applications.
  • Use of open-source frameworks and vector databases.
  • Development of scalable AI assistants for real-world platforms.

7. Duration & Location

  • Duration: 6 months (starting January 2026). The latest start date is March/April 2026.
  • Location: Université Paris-Est Créteil, LISSI Laboratory, 122 rue Paul Armangot, 94400 Vitry-sur-Seine.

8. Application

Please submit:

  • CV
  • Academic transcripts
  • Cover letter
  • Recommendation letters

Send applications to recrutement@declic.net before October 30, 2025.

Notes:

Potential for transition to a permanent position or PhD scholarship.

The selected candidate will join an interdisciplinary team.

Competitive salary aligned with skills and expertise.

Les commentaires sont clos.