✦Agentic AI Engineer Intern @ TransRe · NYC

Hi, I'm Bimal.

I build intelligent systems — from production AI agents to full-stack apps — and I love making technology feel delightfully human.

View my work ↓Resume ↗GitHub ↗
$600K/mo
multi-cloud spend tracked
~700
employees on my RAG assistant
91.7%
recall@5 · 222-question eval
Bimal Kumal in a navy suit on a crosswalk in front of the Oculus in Lower Manhattan
📍 New York City

About me

Curious builder,
forever learning.

I'm a Computer Science & Data Science student at St. Joseph's University, New York, passionate about building production-grade AI systems and beautiful software. I focus on the intersection of machine learning and practical engineering.

Right now I'm an Agentic AI Engineer Intern at TransRe, where I build agents that do real work — a FinOps agent watching $600K/month in multi-cloud spend, and a RAG helpdesk assistant for ~700 employees — and I architected the internal AI platform they run on, on Azure Kubernetes behind a single Okta SSO front door. When I'm not building agents or debugging edge cases, I'm exploring new frameworks and tutoring fellow students in algorithms and programming.

I'm seeking opportunities in AI/ML engineering, full-stack development, and intelligent systems — where I can ship real things that matter.

Claude APIMCPLangGraphRAGPythonFastAPIReactKubernetesAzureAWS
$600KMonthly multi-cloud spend tracked
$115KMonthly recurring savings found
3.95GPA · CS & Data Science
6×Dean's List — top 5% of class

Projects

Things I've built.

Seven projects spanning multi-agent AI, full-stack platforms, deep learning, and statistical ML — each solving a real problem, most of them running in production.

📈● Multi-Agent · FinTech

Arthvion — AI Due Diligence Platform

A multi-agent research platform for private equity that pulls SEC filings, market comps, and risk flags into a full investment memo in under 4 minutes — replacing days of manual analysis. Cut inference costs ~90% and accelerated responses 5× with parallel API calls, exponential backoff, and rolling prompt-cache breakpoints. Shipped with Redis-backed SSE streaming, an atomic SQL ledger for precise billing, rate limiting, and non-root containers.

PythonFastAPIPostgreSQLClaude APIMCPNext.jsDockerTerraform
⚡● AI Agent · Automation

Pulse — Personal AI Agent

An autonomous agent chaining 11 custom tools across 10+ APIs to auto-rank ~70 job postings a week and generate tailored PDF cover letters — cutting application time from ~30 minutes to under 1 minute and saving ~5 hours weekly. Runs 7 daily APScheduler pipelines for AI digests and job alerts, with the public deployment secured by OTP auth, server-side token validation, and API rate limiting.

FastAPIReactOpenAIAPSchedulerREST APIsSQLite
GitHub →Live ↗
🌐● Full-Stack · Cloud

Stellanet — AI Outreach Platform

A full-stack outreach platform that analyzes a student’s research interests to instantly generate ranked professor matches and personalized communication workflows. Amazon Nova is paired with OpenAlex data under a strict-JSON contract to eliminate hallucinations, and parallel university fan-out with caching cut match latency by 4×.

PythonFastAPIPostgreSQLAWS NovaApp RunnerDockerVercel
Live ↗
🧠● RAG · LLM

LLM Document Intelligence

Architected a Retrieval-Augmented Generation pipeline for intelligent document search. Reduced retrieval time by 80% and improved search relevance by 30% using semantic embeddings (FAISS + Chroma). Deployed on AWS with a scalable vector store backend.

LangChainFAISSChromaOpenAI APIAWS
GitHub →
🌸● Deep Learning

Flower Image Classifier

Image classification model using pretrained ResNet18 with transfer learning, data augmentation, and fine-tuning. Built a complete deep learning pipeline from preprocessing through evaluation.

PyTorchResNet18Transfer Learning
GitHub →
🫁● ML · Statistical Analysis

Lung Cancer Prediction

Classification model with Chi-square and Kruskal-Wallis statistical testing. Optimized feature selection pipeline for improved model precision using Random Forest.

RRandom ForestStatistical Analysis
GitHub →
🔥● ML · Regression

Calories Burned Prediction

Achieved an R² score of 0.96 with full data preprocessing and feature engineering. Built a complete ML pipeline from raw data to production-ready evaluation using Linear Regression.

PythonLinear RegressionFeature Engineering
GitHub →

Skills

My toolkit.

Languages, frameworks, and tools I work with daily.

Languages
Python
Java
TypeScript
JavaScript
SQL
R
AI & Agents
Claude API
MCP
Multi-Agent Systems
LangGraph
RAG
Tool Calling
LangChain
OpenAI API
PyTorch
Scikit-learn
Engineering
REST APIs
Real-Time (SSE)
Celery / Redis
CI/CD
Git
Debugging & Testing
Web Frameworks
React
Next.js
FastAPI
Tailwind CSS
Vercel
Cloud & DevOps
Kubernetes / AKS
Azure
AWS Bedrock
AWS App Runner
Docker
Terraform
Istio
Data & Infra
PostgreSQL
pgvector
Redis
SQLite
Microsoft Fabric
Pandas / NumPy
Math & CS
Linear Algebra
Probability
Algorithms
OS Concepts

Background

Where I've been.

Professional, academic, and teaching experience.

Experience

June 2026 – Present
Agentic AI Engineer Intern
TransRe · New York City, NY
  • Built a production FinOps AI agent (Claude tool-use loop + custom FastMCP servers) tracking $600K/month in multi-cloud spend — surfacing $150K in one-time infrastructure rightsizing and $115K/month in recurring savings by reclaiming 400+ inactive M365 licenses
  • Replaced a 3–4 day manual financial reporting process with zero-touch delivery, using Kubernetes CronJobs and a hybrid live-API/Fabric-snapshot pipeline that also eliminated recurring Azure rate-limit failures
  • Architected TransRe's internal AI platform on Azure Kubernetes, replacing per-agent gateways with a single Okta SSO front door at the Istio edge — cutting new-agent onboarding to a one-line database update with zero code changes
  • Launched a RAG IT Helpdesk assistant serving ~700 employees, with a hybrid pgvector/full-text retrieval pipeline validated at 91.7% recall@5 on a 222-question evaluation harness and auto-escalation of unresolved tickets to Freshservice

Education

Expected May 2027
B.S. Computer Science & Data Science
St. Joseph's University, New York
  • GPA 3.948
  • Dean's List (6 consecutive semesters) — top 5% of students
  • Relevant: Data Structures & Algorithms, Advanced Programming, Operating Systems, Database Systems, Software Engineering, Linear Algebra, Calculus I & II, Probability & Statistics

Teaching & Awards

2023 – Present
Honors Peer Tutor
St. Joseph's University
  • One-on-one and group tutoring in pre-calculus & intro CS
  • Created tailored study plans → 10% improvement in test scores
2023
Esse Non Videri Scholar
Scholarship Award
  • Awarded 90% tuition scholarship for academic merit & leadership
Stanford Online
Supervised Machine Learning
Andrew Ng — Coursera
  • Completed Stanford's supervised ML course by Andrew Ng
👋

Get in touch

Let's build something together.

I'm open to AI/ML and software engineering internships, new-grad roles for 2027, and interesting project collaborations. If you think we'd work well together — reach out!

✉️ bkumal@student.sjny.eduLinkedInGitHub📄 Resume (PDF)