Selected Projects

A full-stack AI research workspace that converts conversations, PDFs, documents, and slides into evolving knowledge maps, evidence-linked reports, and reusable research knowledge bases.

• Built LLM pipelines for entity and relation extraction, semantic deduplication, section clustering, and incremental graph updates.
• Implemented source-grounded RAG with evidence lists, hybrid retrieval, reranking, section notes, and cross-section bridge notes.
• Separated immediate LLM answers from asynchronous graph and report construction to keep the product responsive.

An end-to-end crypto reliability scoring service that aggregates multi-source signals such as market integrity, development activity, and on-chain security into a calibrated 0-100 score.

• Built a multi-agent style scoring pipeline from data fetch and schema normalization to subscore inference and weighted aggregation.
• Created an API-backed user experience for real-time scoring with fallbacks for non-EVM assets and infrastructure edge cases.
• Won 3rd place at HackNCState in the Agency Track.

A machine learning framework for predicting stress-strain curves from triblock copolymer parameters, with forward and inverse models supporting material performance exploration.

• Predicted stress-strain curves from polymer structure-related parameters.
• Integrated Bayesian optimization to support process and performance improvement.
• Connected simulation parameters with physical behavior in a more interpretable workflow.

A convolutional autoencoder approach for analyzing and reconstructing twin screw extruder elements, supported by PCA for encoded-feature visualization.

• Applied machine learning techniques to monitor and optimize extrusion processes.
• Built representation learning workflows for screw element analysis and reconstruction.
• Used PCA to reveal patterns in encoded features.

A collection of game AI experiments spanning movement behaviors, classic path finding, and reinforcement-learning-driven agents in interactive environments.

• Implemented AI behaviors such as orientation and evade strategies.
• Built Dijkstra and A* path-finding workflows for navigation problems.
• Explored multi-agent behaviors and RL in game scenarios.

A Near-Earth Object analysis workflow combining NASA JPL data and close-approach API data to study object characteristics, cluster patterns, and classification performance.

• Merged CSV datasets with live NASA data, cleaned missing values, and normalized numerical features.
• Performed EDA, K-Means clustering, KNN classification, and PCA visualization.
• Used the project as a compact end-to-end ML analysis case study.