I'm primarily an AI engineer, building systems that retrieve, reason, predict, and turn complex data into useful decisions. I'm also a full-stack software engineer who carries those ideas from models and APIs to reliable interfaces people can actually use. Before choosing a model or framework, I focus on the why behind the problem—who experiences it, what is getting in their way, and what would make the solution genuinely useful. I care about the path from an interesting model to a dependable product.

Languages & product

Backend & systems

AI & machine learning

Five problems, five working systems—each built from the reason it needed to exist.

• Situation: Designing a dependable RAG pipeline meant repeatedly testing disconnected choices for chunking, embeddings, retrieval, reranking, quality, and cost.
• Task: Create one workspace where developers could build, compare, evaluate, and export complete retrieval pipelines.
• Action: Built a Next.js and FastAPI platform with a drag-and-drop pipeline editor, AI configuration suggestions, multiple retrieval strategies, pgvector storage, live RAGAS-style evaluation, cost estimation, and Python code export.
• Result: Delivered an end-to-end environment that makes RAG experiments visible, comparable, and reusable instead of leaving decisions scattered across notebooks and scripts.

• Situation: Training a useful ML model often requires repetitive setup and enough technical knowledge to choose algorithms, evaluate results, and interpret metrics.
• Task: Make structured machine-learning experimentation approachable without hiding the reasoning behind the output.
• Action: Built a guided workflow for uploading datasets, selecting algorithms, training models, and reviewing evaluation metrics, visualizations, and plain-language explanations.
• Result: Created a single workspace that takes an experiment from raw dataset to an understandable trained model, making iteration faster and more accessible.

• Situation: Online examinations need more than question delivery: institutions must coordinate candidates, examiners, administrators, integrity checks, grading, and live oversight.
• Task: Lead the development of a unified, role-based platform that could support the complete examination lifecycle.
• Action: Engineered student, examiner, and admin workflows; added browser-side MediaPipe face and gaze detection, fullscreen controls, live violation monitoring, analytics, multilingual support, and heuristic grading designed for later LLM replacement.
• Result: Delivered a full-stack, installable examination system that brings exam delivery, proctoring, evaluation, and administrative monitoring into one product.

• Situation: Students often track their mood, habits, reflections, and wellbeing across disconnected tools—or stop tracking them altogether.
• Task: Design a calmer, private space that could help students understand daily patterns and keep making progress.
• Action: Built a unified experience for mood check-ins, habit building, private journaling, and guided reflection, with an interface designed around everyday student life.
• Result: Turned several fragmented wellbeing routines into one coherent workspace that supports reflection without adding more friction.

• Situation: Recognizing plant disease from visible symptoms can be difficult for users without immediate access to specialist guidance.
• Task: Build a simple path from a plant image to an actionable first assessment.
• Action: Developed an image-upload workflow backed by computer-vision classification, then connected predictions to disease information and treatment recommendations in a focused dashboard.
• Result: Created an accessible screening experience that converts a leaf image into a clear, useful next step for plant care.

Jun 2026 — Aug 2026 · Remote

• Led a team building a full-stack AI-proctored examination platform for student, examiner, and admin workflows.
• Implemented client-side face detection with MediaPipe, automated grading designed for future LLM replacement, live monitoring, and attempt analytics.

Aug 2025 — Oct 2025

• Contributed 20+ pull requests across five Python projects and improved work through review feedback.
• Ranked among the top 2% of 30,000+ contributors nationwide.