Graduating May 2027 · Rutgers Computer Science

Pratham Bhatt

CS student at Rutgers building LLM agents and publishing research in cloud computing and defect prediction.

I'm a senior at Rutgers University–New Brunswick, graduating in May 2027 with a B.S. in Computer Science. I build LLM-powered tools, have interned in data/AI engineering, QA and business analysis, and have two conference papers accepted in 2026.

Resume LinkedIn GitHub

1,000person hackathon enabled by my SkillsProject work
2conference papers accepted in 2026
99.15%best accuracy in my defect-prediction paper
3internships across AI, business and QA

Experience

Next chapter

Still building, still learning.

Data/AI Engineer Intern

September 2025 – April 2026

SkillsProject

Built LLM-driven synthetic dataset curation for the platform's dynamic micro-challenge generation using the Claude API, and validated role-specific technical challenges — the core project that enabled a 1,000-person hackathon.
Implemented quality-checking and difficulty calibration logic so generated challenges matched candidate skill level and assessment standards.
Developed front-end features and interactive coding editors for the candidate assessment portal.
Built fraud-detection heuristics to identify anomalous candidate behavior and preserve assessment integrity.
Wrote provisioning scripts to spin up containerized assessment environments and supported CI/CD workflows to help scale infrastructure for high-volume events.

Business Analyst Intern

July 2025 – September 2025

Mainstream Tek

Independently completed structured SAP training via the SAP Learning Hub, building foundational proficiency in ERP concepts and modules used to support core business functions.
Self-directed research into how enterprise systems like SAP enable data-driven decision-making, connecting technical platform capabilities to real business analyst workflows.

Quality Assurance Intern

July 2022 – August 2022

United Flow Technologies

Assisted in QA of their public website, identifying defects like broken web links, incorrect grammar/spelling, and miscellaneous bugs, reported using JIRA.

Projects

Upwork Job-Matching Agent

2026

Python
Claude API
Streamlit

Built a multi-agent system that automates the job-search workflow, with specialized agents for job matching, cover-letter generation, and interview prep, coordinated by an orchestrator agent.
Designed agent tools, function-calling logic, and inter-agent coordination directly on Anthropic's Claude API, modeling the architecture on established agent frameworks.
Developed in Claude Code and evaluated a Streamlit UI layer to make it deployable as a usable app.

View on GitHub →

AI Data Analysis Agent

2026

Python
Claude API
SQL

Built an agent that lets users query datasets (CSV, XLSX, TSV, SQLite) in plain English by translating questions into SQL.
Implemented a two-tool loop (dataset inspection + query execution) with error handling and self-correction so the agent recovers from failed queries automatically.

View on GitHub →

Publications & Research

Accepted and presented · 2026

AI-Driven Defect Prediction and Risk Analysis Framework Based on SDLC Platforms

A. Baldwa, P. Bhatt, D. Upadhyay, and P. Bhatt

International Conference on Emerging Trends in Networks and Computer Communications (ETNCC 2026)

Merges two public defect datasets (49,498 records) and compares Extra Trees, XGBoost, MLP and LSTM models. The MLP reached 99.15% accuracy and 99.79% ROC AUC, and SHAP analysis surfaces the features that drive defects.

Accepted and presented · 2026

Adaptive Workload Scheduling for Efficient Resource Utilization in Cloud-Native Orchestration Systems

P. Bhatt, A. Baldwa, and P. Bhatt

1st International Conference on Statistical Learning, Data Science & Generative AI (ICSL-DSGA 2026)

AWSO, an adaptive scheduler for cloud-native orchestration. It beat the best baseline scheduler, cutting scheduling latency by 27–50 s and lifting SLA compliance to 95.6–96.4% across small, medium and large workloads.

Previews show the accepted manuscripts with author details omitted. Full papers will be linked once published.