Hi, I'm Jay Kalbi 👋

Building production-grade Large Language Model systems, Financial AI, Multi-Agent Swarms, Enterprise RAG, and Edge AI solutions.

I am an AI Engineer specializing in Large Language Models (LLMs), Multi-Agent Systems, Explainable AI (XAI), and Predictive Analytics for high-stakes enterprise applications.

Currently pursuing my M.Tech in Artificial Intelligence at NMIMS MPSTME, I focus on architecting hybrid ML/LLM systems, regulatory-compliant AI risk engines, multi-stage RAG pipelines, and edge-deployed computer vision systems backed by government grants.

• 🎓 Education: M.Tech in AI @ NMIMS MPSTME (2026–Present) | B.Tech in Computer Engineering @ BVM (CGPI: 8.84)
• 💼 Experience: Ex-Machine Learning Intern @ HPParam IT Solutions | Ex-Backend Intern @ TechnoGuide Infosoft
• 🎯 Current Focus: Building BankOS Enterprise Banking Platform, Multi-Agent AI Systems, Cloud-Native Architecture, Agentic RAG, Financial AI, Platform Engineering, MLOps, and Quantized LLM Fine-tuning (QLoRA).

A production-grade, enterprise-scale banking platform reference implementation designed to mirror the engineering practices, cloud architecture, security, and platform capabilities of modern Fortune 100 financial institutions.

• 🏛️ Enterprise Banking Platform: Building a distributed, event-driven banking operating system with domain-driven microservices, API-first design, zero-trust security, and GitOps-managed cloud infrastructure.

• ☁️ Cloud-Native Platform Engineering: Architecting scalable services on AWS EKS (Kubernetes) and AWS Fargate, leveraging Infrastructure as Code, GitOps workflows, and automated deployment pipelines.

• ⚡ Enterprise Infrastructure: Integrating PostgreSQL, Kafka, Redis, MinIO, Milvus, and Snowflake to support transactional workloads, event streaming, vector search, object storage, and enterprise analytics.

• 🔐 Security & DevSecOps: Implementing Zero Trust architecture with HashiCorp Vault, SOPS, Trivy, Cosign, Gitleaks, and supply-chain security following modern DevSecOps best practices.

• 📊 Observability & Reliability: Building enterprise monitoring with OpenTelemetry, Prometheus, Grafana, and Jaeger for distributed tracing, metrics, logging, and production diagnostics.

• 🤖 AI-Ready Banking Platform: Engineering a modular platform capable of supporting future AI-powered banking services including intelligent assistants, fraud detection, credit risk analytics, document intelligence, regulatory AI, and autonomous financial workflows.

• 🛠️ Tech Stack: Java 21 Python 3.11 Node.js 20 Go AWS EKS Kubernetes AWS Fargate PostgreSQL Kafka Redis Snowflake Milvus MinIO Terraform Crossplane Helm ArgoCD OpenTelemetry Prometheus Grafana Jaeger Vault Trivy Cosign

• 🚧 Status: Enterprise architecture completed. Platform engineering foundation currently under active development.

• 🔗 Repository: github.com/JayKalbi/BankOS

An institutional multimodal credit risk assessment platform fusing quantitative gradient boosting (LightGBM) with fine-tuned Mistral-7B LLMs and autonomous multi-agent underwriting swarms.

• ⚙️ Multi-Agent Committee Swarm: Implemented a 4-agent committee swarm (Quant Auditor, Macro Strategist, Compliance Officer, CRO) to debate credit limits, integrate Vasicek CCAR macro stress shocks (Fed Rate/Unemployment), and enforce CFPB 12 CFR § 1026.43 Ability-to-Repay legal citations.
• 📊 Performance Metrics: Achieved 0.9845 AUC-ROC vs. 0.6709 baseline across 415,000+ HMDA mortgage applications. Includes W-2 / Tax 1040 Income Fraud Audit Engine.
• ⚡ MLOps & H100 Optimization: Containerized microservices via Docker & Render; optimized NVIDIA H100 GPU fine-tuning (QLoRA $r=64$, bfloat16, vLLM tensor parallelism) with a 44.5 GB VRAM peak allocation.
• ⚖️ Compliance & XAI: Enforced 100% ECOA Fair Lending 80% Rule compliance and SHAP game-theoretic explainability, preventing disparate impact across protected demographic cohorts.
• 🛠️ Tech Stack: Python PyTorch Hugging Face QLoRA (r=64) bfloat16 vLLM LightGBM SHAP Flask Docker Render NVIDIA H100
• 🌐 Live Demo: institutional-risk-engine.onrender.com
• 🔗 Repository: github.com/JayKalbi/institutional-risk-engine

Production-hardened hybrid retrieval-augmented generation system for SEC 10-K financial document question-answering.

• ⚙️ Retrieval Pipeline: Multi-stage retrieval combining Dense Vector Search (all-MiniLM-L6-v2 in ChromaDB) and Sparse Search (BM25 Okapi) merged via Reciprocal Rank Fusion (RRF), followed by a Cross-Encoder Reranker (ms-marco-MiniLM-L-6-v2).
• 🧠 Grounded Generation: Powered by Llama 3.1 8B Instant (Groq) with automated inline citation validation to eliminate hallucinations. Automated LLM-as-Judge evaluation using Llama 3.3 70B.
• 🛠️ Tech Stack: Python LangChain ChromaDB BM25 Okapi Cross-Encoder Llama 3.1 8B Llama 3.3 70B Flask Waitress Docker GitHub Actions
• 🔗 Repository: github.com/JayKalbi/hybrid-rag-engine

Award-winning real-time computer vision & automation platform deployed on edge hardware.

• 🏅 Grant & IP: Awarded ₹1,20,000 grant under Student Startup & Innovation Policy (SSIP), Govt. of Gujarat. Pending IPR Patent.
• ⚡ Optimization: JetPack 4.6 on NVIDIA Jetson Nano/Orin with custom swapfile memory allocation, CUDA pipelines, and RTSP video streaming, achieving +25% cost efficiency.
• 🛠️ Tech Stack: NVIDIA Jetson CUDA OpenCV MediaPipe TensorFlow Python RTSP
• 🔗 Repository: github.com/JayKalbi/JetsonNano-OpenCV-Mediapipe-SetUP

Intelligent text classification pipeline built for enterprise email workflow automation.

• 📩 Performance: Trained Bidirectional LSTM neural network on 500K+ records (Enron Email Dataset) achieving ~90% classification accuracy across Business, HR, Legal, and Finance categories.
• 🔬 Analytics: Preprocessing with KNN clustering for label reduction, PCA & t-SNE embedding visualization, Gmail API integration, and an interactive Streamlit dashboard.
• 🛠️ Tech Stack: TensorFlow/Keras BiLSTM NLP Scikit-Learn t-SNE PCA Gmail API Streamlit Pandas
• 🔗 Repository: github.com/JayKalbi/Email-Classification-with-LSTM

• 💼 LinkedIn: linkedin.com/in/jaykalbi
• 🌐 Portfolio: jaydkalbi2004.wixsite.com/portfolio
• 📧 Email: jdkalbi18@gmail.com

"Building intelligent, production-ready AI systems that drive real-world impact."