Enhancing planning and constraint satisfaction capabilities…

I am a postgraduate student in the MSc in Artificial Intelligence and Deep Learning (University of West Attica) and my thesis is titled:

Enhancing Planning and Constraint Satisfaction Capabilities in Large Language Model Agents: An Experimental Study Using the TravelPlanner Benchmark.

The thesis is experimental and implementation-oriented and is based on the official TravelPlanner repository (

).

The main objectives are:

Evaluation of a single-agent baseline on the benchmark.

Design and implementation of a multi-agent workflow (e.g., planner, verifier, refinement loop).

Systematic and fair experimental comparison between single-agent and multi-agent approaches.

Formulation of a clear experimental matrix (models policies budgets seeds metrics).

Statistical documentation of results (multiple seeds, mean std and, where required, significance testing).

Limited ablation studies and failure analysis.

Investigation of the possibility of improving performance compared to the benchmark baseline results, under controlled and fair experimental conditions.

Extension to multi-agent orchestration through a ready-made drag-and-drop tool (e.g., Waldiez framework).

Important technical clarification:

The implementation will be based exclusively on open or freely accessible LLM models (e.g., NVIDIA-hosted open models or similar). I do not wish to use commercial APIs with per-call charges.

Additionally, in the official repository the postprocess/parsing (conversion of the natural language plan into structured JSON) is stated to potentially use GPT-4. I would like to agree from the outset on a no-paid-API mode, meaning:

Either replacing the parsing component with an open/free LLM,

Or implementing deterministic/constrained JSON parsing,

so that no commercial API key is required at any stage (agent run, postprocess, or evaluation).

Proposed Hybrid Collaboration Model:

I am not seeking full outsourcing of the thesis, but rather collaboration with a clear division of responsibilities.

Phase 1 Initial Technical Boost

Mapping and proper setup of the repository

Configuration of the experimental matrix

Design of the multi-agent architecture

Creation of a minimal reproducible pipeline (validation subset)

Configuration of no-paid-API mode

Structuring of the methodology chapter

Phase 2 Collaborative Implementation

I will personally undertake the execution of experiments and basic parameter tuning

You will provide scientific review of the results

Guidance on statistical analysis

Support in writing the technical chapters

Support for revisions following supervisor feedback, within the agreed scope

For clarity, I would appreciate if the offer includes:

A total fixed cost for the above hybrid model

A clear description of deliverables per phase

An indicative timeline with checkpoints

Explicit reference to the number of revision cycles included

A clear definition of what is considered outside the agreed scope

Confirmation that there will be no additional charges for in-scope adjustments

Delivery of fully editable code with execution instructions (README)

Clarification of the academic background of the collaborator who will provide the technical guidance

I would like to ensure that the collaboration framework will be genuinely supportive and interactive, as I wish to have active participation and full understanding of the implementation. Specifically, I would like to confirm that:

The code will be delivered in fully editable form (structured repository with README and execution instructions).

I will be able to run and modify the code throughout the collaboration.

There will be intermediate deliveries (e.g., Phase 1 setup) before final completion.

Specific revision cycles following supervisor feedback will be included within the agreed scope.

At least 12 technical online sessions will take place for a walkthrough of the code and experimental design.

The implementation will function fully without the use of commercial per-call APIs (no-paid-API mode), as stated above.

There will be no additional charges for adjustments that fall within the agreed experimental framework.