You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
I am the sole author of “Hyperparameter Optimisation of Convex Portfolio Trajectories using Large Language Models,” which has been accepted for presentation at the EvoRobust workshop at NeurIPS 2026 in Sydney.
The research explores the use of LLMs to guide hyperparameter optimisation in constrained, multi-period convex optimisation. I am now extending this work toward a more general question: how reliable are LLM-guided optimisation systems when their specifications, constraints, inputs, or environments change?
I am requesting $2,500 in funding to attend NeurIPS in person, present the accepted research, and utilise the conference to develop this next stage of the research with researchers working on LLM evaluation, robustness, optimisation, and AI safety.
The immediate goal is to present the accepted paper at NeurIPS and receive technical feedback from researchers working in machine learning and optimisation.
The next research goal is to systematically evaluate failure modes of LLM-guided optimisation systems, including sensitivity to specification changes, constraint perturbations, distribution shift, repeated-run instability, and pathological inputs. The existing portfolio-optimisation framework provides a useful controlled environment because the objectives, constraints, and performance metrics are explicit and quantitatively measurable.
I will use feedback and discussions at NeurIPS, as well as the broader Sydney AI safety community, to refine the evaluation methodology and identify the most useful experiments for the next iteration of the work.
The $2,500 will contribute directly to the cost of attending NeurIPS 2026 in Sydney, primarily airfare, accommodation, registration, and local transportation.
Attending in person is important because the research is at a stage where technical feedback from researchers working on LLM evaluation, robustness, optimisation, and AI safety can directly shape the next iteration of the project.
I am currently the sole researcher on this project.
I am entering the UC Berkeley Haas Master of Financial Engineering program and have a background spanning mathematical optimisation, quantitative finance, machine learning, and statistical modelling. I graduated from IIT Madras with a Department Rank of 2/115 and have worked in quantitative roles at Fidelity Investments and BlackRock.
My research has previously been accepted at a NeurIPS 2025 workshop and an ICML 2026 workshop, and I have also presented research at ICML in person. My work has included convex and linear optimisation, Gurobi, Monte Carlo simulation, statistical modelling, portfolio construction, and production machine-learning systems.
I have also previously received a $2,000 BlueDot Impact Rapid Grant supporting my transition toward technical AI safety research. I successfully completed the objectives specified in that grant, including presenting my accepted ICML 2026 workshop research.
The most likely failure mode is not that the underlying research stops, but that I am unable to attend NeurIPS in person or participate fully due to the cost of international travel.
In that case, I would continue developing the research independently, but I would lose the opportunity to present the work in person and receive concentrated technical feedback from researchers working on robustness, evaluation, optimisation, and AI safety.
The research could therefore proceed more slowly and with less external feedback. If the proposed robustness extension itself does not yield useful results, I would still have an accepted research contribution and be able to use the negative results to identify which approaches to LLM-guided optimisation are unreliable and why.
I received a $2,000 BlueDot Impact Rapid Grant in 2026 to support presenting my accepted ICML 2026 workshop research while transitioning toward technical AI safety research. I successfully achieved the objectives specified in that grant
I have not otherwise raised dedicated external funding for this NeurIPS project.