>I build AI systems that ship.

AI Engineer Intern · Transformly AI

Sole engineer on the 0-to-1 MVP: a multi-agent platform that automates ML workflows from raw data to ranked model recommendations.

• Architected a LangGraph platform of 3 orchestration graphs and 5 modeling subgraphs covering ingestion, EDA, model training, ranking and feature-importance analysis, adopted by 2 customer companies.
• Partnered directly with the CEO to scope product priorities, translating shifting customer needs (e.g., adding time-series forecasting for a client) into build decisions, validating direction before committing engineering time.
• Built an LLM project discovery and planning pipeline that reads business context, EDA output and dataset schemas to generate and rank up to 15 ML use cases by ROI and feasibility, cutting initial project planning from hours to under 4 minutes.
• Engineered 60+ ML tool modules and 25+ human-in-the-loop checkpoints on LangGraph conditional routing and persistent state, making long workflows modular, auditable and resumable.
• Led the AWS migration: 3 autoscaling Dockerized services (backend, UI and worker) on ECS Fargate with ECR, S3, SQS/DLQ, ALB, IAM, CloudWatch, Secrets Manager and Terraform.
• Designed an LLM eval framework in LangSmith and DeepEval covering 18 production-critical calls, spanning prompt regression, hallucination detection, JSON schema validation and token cost profiling. Caught 4 regressions pre-deploy.
• Diagnosed a production bug where workflows resumed from stale checkpoints after human-in-the-loop decisions, tracing it via log analysis to per-node dataset path tracking that didn't update correctly; fixed it by making state updates synchronous after every step.

Teaching Assistant · CS1800 Discrete Structures · Northeastern University

TA for a ~300-student undergraduate discrete structures course.

Teaching Assistant · CS3000 Algorithms and Data · Northeastern University

TA for a ~400-student undergraduate algorithms course covering sorting, searching, graph algorithms, dynamic programming, linear programming and computational complexity.

• Held weekly office hours, helping students work through algorithmic problem-solving and assignment debugging.
• Graded assignments and exams for a ~400-student class, maintaining timely feedback and consistent evaluation standards.

AI Research Intern · MySellerCentral

Retrieval and guardrail work on a live seller-listing pipeline handling tens of thousands of listings a month.

• Built a hybrid RAG pipeline (FAISS, sentence-transformers, LangChain, GPT-4o) that lifted HSN code classification accuracy from 72% to 94% while cutting inference cost per classification roughly in half.
• Shipped a multimodal LLM-as-a-judge validation pass with targeted regeneration, keeping listings factually grounded in seller input. Auto-approval rates rose 12% and human review overhead fell.
• Implemented prompt-level PII guardrails protecting 30K+ monthly listings, ensuring compliance with data protection policies; deployed as a Docker microservice on AWS Lambda with GitHub Actions CI/CD.

Software Engineering Intern · Hoonartek

Built CostSentry, a cloud cost governance platform for Snowflake users.

• Cut manual cost tracking by 70% and improved spend oversight across 1500+ enterprise accounts.
• Optimized SQL pipelines against Snowflake Data Cloud, dropping query latency from ~11s to 5s to make the cost dashboards near real-time.