The Grenoble Computer Science Laboratory (Laboratoire d’Informatique de Grenoble) in Grenoble, France, is opening a 12-month (extensible one time up to 24-month period in total) post-doctoral research fellowship position starting from September 2024 in the fields of **Large Language Models (LLMs)**, **Information Systems (ISes),** and **Low-Code Development Platforms (LCDPs)**.
Title: DRaw me A French information sysTem (DRAFT)
## Context
With the increased value of information, developing Software tools to manipulate, store, and analyse it has become a challenging task. The ability to quickly take decisions based on the collection of heterogeneous, possibly large amount of data originating from different sources in real-time requires complex representations and advanced analysis tools that are now more and more integrated in large information systems, making their development even more complex. In addition, due to the market pressure, companies must reduce production cycle times. This has a negative impact on developers who have to constantly innovate to develop software applications faster while not compromising robustness.
The DRAFT project intends to address this challenge by combining two paradigms, namely low-code development platforms (LCDPs) and large language models (LLMs). On the one hand, LCDPs promote the development of of fully-functional and customizable applications, using visual abstractions and graphical user interfaces while requiring a small amount of code. On the other hand, the recent outbreak in Research and development of LLMs open new perspectives in terms of software production from textual specifications. Yet, as pointed out several times in the literature, LLMs often produce mitigated results when used to generate code, forcing developers to spend time reviewing the code generated by the LLMs, thus offsetting the possible benefits of using them. One reason that can explain the lack of performance from LLMs is the high expressiveness general-purpose programming languages such as Java can offer. This limitation is however counterbalanced in the context of low-code development as the expressiveness a LCDP can offer is often curbed by the number of available cots-of-the-shelf widgets, along with their customization capabilities. The main objective of the DRAFT project is to develop a smart no-code platform where end users are able to express the specifications of information systems (ISes) in high-level abstraction grammars --- also known as Domain Specific Languages (DSLs).
The project involves researchers from two teams from the LIG laboratory, one (VASCO) specialized in the modeling, verification and validation of systems, and one (GETALP) specialized in natural language processing. In the context of the DRAFT project, a partnership is currently under discussion with French local authorities to define a case study. It brings one technical constraint in the project. Since the end users are French native speakers, the LLMs that will be used during the project must be trained over French specifications.
## Duties
The postdoctoral fellow’s work plan will be as follows :
**Literature Review and Research (T0-T3):**
1. Conduct in-depth reviews of generative AI, particularly open-source Large Language Models (LLMs) ;
2. Explore existing low-code development platforms, identifying strengths, weaknesses, and opportunities for integration ;
3. Engage in monthly interviews with end-users, to understand their software environments and tool requirements.
**Low-Code Platform Development (T3-T6):**
1. Familiarize himself or herself with the existing low-code platform developed at the LIG laboratory ;
2. Contribute to the development of the low-code platform, integrating an open LLM to generate both data models and user interfaces ;
3. Collaborate with the team to enhance the structural and presentational aspects of the information.
**Integration and Experimentation (T6-T18):**
1. Integrate and experiment with the LLMs, focusing on training them for optimal performance ;
2. Collaborate with the Pantagruel LLM team to exchange open-source code, data, and best practices ;
3. Generate interfaces aligned with user expectations, addressing challenges associated with data model generation and interface development.
The scientific results are expected to be published in in international journals and conferences. Relevant conferences and journals include *ACL, LREC, COLING, MODELS, SoSyM, etc.
## Work Environment
The Grenoble Computer Science Laboratory offers a dynamic and collaborative research environment. The postdoctoral fellow will have access to computational resources, including those of the LIG and French supercomputers. The laboratory is part of MIAI, which has recently received the French government’s IA cluster label, providing an excellent work environment with cutting-edge facilities and opportunities for interdisciplinary collaboration.
## Qualification
The selected candidate must have a PhD degree in Computer Science or an equivalent degree with a strong focus on at least one of the two areas LLMs or LCDPs. We expect the student to be motivated, innovative, independent without close supervision, and interested in upskilling in the other area. Although it is expected work on a French LLM at some point in the project, it is not expected to be fluent in French and non-french speaker candidate are welcomed to apply.
## Procedure
Please send the following documents to Nicolas Hili <nicolas.hili@univ-grenoble-alpes.fr> and Didier Schwab <didier.schwab@univ-grenoble-alpes.fr>
- Your CV ;
- Your PhD’s degree certificate and any publications relevant to the proposed position ;
- At least one letter of recommendation.