Hey, I'm
Turning AI/ML models into working systems
I'm a third-year computer science student who enjoys building things and understanding how they work beyond the surface.
I learn by building projects, breaking ideas down, fixing mistakes, and exploring applied AI and machine learning systems in real-world contexts.
View ResumeProjects
A system that removes friction from meetings.
Meetings create information, but decisions and ownership get lost.
Transform audio into structured decisions and actionable tasks.
Speech → Text → schema-validated task extraction.
Clear decisions, assigned tasks, instant results, reduced manual effort.
Built with Flask, PostgreSQL, Deepgram and Gemini
Giving stateless AI assistants long-term conversational memory.
LLMs are stateless and cannot remember past conversations.
Add a persistent memory layer that stores and retrieves conversations using embeddings.
Messages → embeddings → vector database → similarity retrieval → injected into the LLM prompt.
AI assistants gain long-term conversational memory and controllable forgetting.
Built with Endee Vector Database, Sentence Transformers, Groq (Llama 3.3)
Understanding users before they leave.
Churn is usually identified after users leave, making it hard to understand which behavioral signals indicated risk earlier.
Model churn risk by analyzing how user activity patterns change over time instead of relying on final churn events alone.
Built a behavioral data pipeline, engineered features, and trained baseline models to estimate early churn risk under practical constraints.
Gained a clearer understanding of how ML systems behave in practice, including feature design, evaluation trade-offs, and system limitations.
Built with Python, PostgreSQL, and Logistic Regression
Contact