Seemanta Bhattacharjee

This project designs and evaluates multi‑agent systems (MAS) that leverage LLM agents with Retrieval‑Augmented Generation (RAG) to generate and debug quantum programs. The assistant provides a conversational interface grounded in the latest Qiskit releases, reducing hallucinations and outdated API usage while helping with technical questions, code development and execution, debugging, and visualization.

Recipient of the 3rd Best Presentation Award at QIntern'25 (QWorld). We cast underground pipe routing as a QUBO that enforces engineering constraints such as valid fitting angles, downhill wastewater flow, and excavation cost. To overcome QAOA's quadratic qubit growth, we introduce a hierarchical decomposition that scales to large graphs. Benchmarks against a classical A* baseline match costs on small networks and characterize the trade‑off on industrial‑scale cases (10,000+ nodes).

We performed a comprehensive study applying quantum kernels to financial time‑series data, using a self‑curated Dhaka Stock Exchange (DSEx) Broad Index dataset. Implemented and benchmarked multiple quantum‐kernelized SVM models, identified the optimal kernel for our dataset, and demonstrated an empirical quantum advantage. To ensure robustness, we validated results against the Phase Space Terrain Ruggedness Index and projected the computational resources required to scale this approach for future practitioners.