open to work available for new opportunities building new things open to work

Pritam Mondal.

SDE II · Backend · Platform · Distributed Systems · AI

Senior software engineer with 8+ years designing and operating backend systems, distributed workflow platforms, industrial IoT streaming, AWS infrastructure, and applied AI systems. Currently at Cornerstone OnDemand.

View my work GitHub ↗ LinkedIn ↗

$590k

Annual savings delivered

11.6×

Backup runtime improved

Years of experience

#25

App Store ranking

pritam@csod ~

›role SDE II @ Cornerstone

›focus CSX Scheduler · Harvest

›stack .NET · AWS · AI

›status ● open_to_work

C# / .NET 8 AWS SQS Quartz.NET Kafka / MSK vLLM LoRA / PEFT Aurora PostgreSQL CloudFormation MCP RAG API Gateway ECS / ECR GovCloud PyTorch DynamoDB C# / .NET 8 AWS SQS Quartz.NET Kafka / MSK vLLM LoRA / PEFT Aurora PostgreSQL CloudFormation MCP RAG API Gateway ECS / ECR GovCloud PyTorch DynamoDB

01 · About

Building systems
that scale and last.

I'm a Software Development Engineer II at Cornerstone OnDemand, promoted in April 2025. With 8+ years across backend, cloud, industrial IoT, and AI systems, I specialize in distributed platforms and applied AI — often in the same project.

My biggest engineering impact so far has been CSX Scheduler — an in-house distributed job orchestration platform I architected from scratch to replace a $700k/year legacy vendor dependency. It now runs at ~$110k/year with better reliability, blue-green deployments, and full AWS-native infrastructure.

At LTIMindtree, I built real-time factory monitoring systems for Toyota — connecting on-prem OPC/Kepware machine telemetry through Kafka/MSK, AWS Glue, SQS/Lambda, API Gateway WebSockets, and DynamoDB to live Andon dashboards with SageMaker-hosted anomaly detection.

I bridge backend engineering, cloud infrastructure, industrial IoT, and AI experimentation naturally — comfortable debugging Quartz triggers, designing FedRAMP-compliant IAM policies, or fine-tuning an LLM on a G5 instance.

        // M.Tech in AI & ML — BITS Pilani (2023–2025)

        // AWS Certified Solutions Architect – Associate

        // Based in Kolkata, India

⚡

$700k → $110k/yr

Replaced JAMS legacy scheduler with CSX Scheduler, saving ~$590k annually in platform operating cost.

🚀

6 hrs → 31 min

Rebuilt backup workflow using controlled concurrency, robocopy & mutex design — 11.6× faster, $40k/yr in avoided EC2 scale-up.

🟢

Zero-downtime deploys

Designed blue-green deployment for CSX Scheduler, reducing 6+ hr maintenance windows to near-zero client impact.

🔒

GovCloud / FedRAMP

Deployed CSX Scheduler into US GovCloud restricted environments, navigating compliance and networking constraints.

🏭

Real-time factory monitoring

Built Toyota's IoT pipeline: OPC/Kepware → Kafka/MSK → AWS Glue → WebSockets → Andon dashboards, with SageMaker Isolation Forest anomaly detection.

📱

Top #25 App Store

Built and shipped BorderApp for iOS — ranked #25 in Photos category, fully offline, no login, no tracking.

02 · Experience

Where I've built
things that matter.

Cornerstone OnDemand Dec 2022 – Present

Software Development Engineer II

Apr 2025 – Present

Designed AWS ALB + Route53 endpoints for CSX Scheduler, replacing fragile direct EC2 hostnames and eliminating ASG-related DNS failures across all environments.
Architected blue-green deployment strategy — reduced client-impacting maintenance windows from 6+ hours to near zero downtime.
Deployed CSX Scheduler to US GovCloud / FedRAMP restricted environments, working through compliance, AMI, service account, and network constraints.
Led Project Harvest — a multi-agent AI document intelligence pipeline for OCR/vision extraction, field mapping, validation, and structured publishing at scale.
Resolved production incidents including stuck jobs after server refreshes, JFrog Artifactory edge cache inconsistencies, and deployment/environment failures.

AWS Route53ALBCloudFormationBlue-GreenGovCloudMulti-agent AIOCR/Vision

Software Development Engineer I

Dec 2022 – Mar 2025

Architected CSX Scheduler from zero — a distributed job orchestration platform replacing JAMS. Cost reduced from $700k → $110k/yr.
Built queue-driven job execution using AWS SQS, Quartz.NET, .NET Background Services, and processor agents with visibility timeout handling and failure recovery.
Optimized feed server backup from 6 hrs to 31 min using controlled concurrency, mutex locking, and robocopy — avoiding $40k/yr in EC2 scale-up.
Introduced C# Channels + jittered backoff to eliminate thundering herd behavior in the processor service under concurrent job polling.
Migrated CSX Scheduler to .NET 8, modernized database layer to Aurora/PostgreSQL, refactored Lambda structures to internal PLK format.
Built MCP servers for CSX Scheduler and FeedStats — enabling AI agents to interact with internal enterprise systems through structured tool interfaces.
Designed, fine-tuned, and deployed CSOD-R1 — an internal MoE-style LLM built on Qwen2.5-7B-Instruct using LoRA/PEFT on AWS G5.2xlarge (NVIDIA A10G), ~$10/training run. Built and trained CSOD-IC intent classifier on DistilBERT; co-hosted CPU-side with GPU vLLM inference for compute efficiency.
Built Pebbles VS Code extension for AI-assisted unit test generation using internal LLM endpoints.
Automated feed tool deployment pipeline across environments, accounts, regions — parallel orchestration with master job, Artifactory integration, and detailed logging.
Drove test coverage above 85% for CSX Scheduler components; created runbooks, architecture diagrams, and wiki documentation.

C# / .NET 8ASP.NET CoreAWS SQSQuartz.NETAurora PostgreSQLCloudFormationJenkinsvLLMMCPRAGPowerShell

LTIMindtree Jan 2019 – Dec 2022

Senior Software Engineer

Jan 2021 – Dec 2022

Built backend and frontend for Toyota PCMS (Production Control Management System) — production planning, vendor management, admin workflows, Andon dashboards. Connected with CDTS for coil data visibility.
Developed Toyota CDTS (Coil Data Tracking System) — real-time coil tracking with near-real-time Andon displays for blanking line operations.
Built Toyota Real-Time Factory Monitoring Platform: On-prem machines → OPC/Kepware → Hangfire .NET ingestion → Kafka/MSK → AWS Glue (transform + RDS PostgreSQL) → SQS → Lambda → API Gateway WebSockets → Andon dashboards. DynamoDB for connection/context routing so updates targeted only relevant clients.
Integrated SageMaker-hosted Isolation Forest anomaly detection — surfaced abnormal KPI behavior as live warnings on Andon dashboards.
Deployed containerized .NET Core APIs via Docker → ECR → ECS, managed through Jenkins + XL Release pipelines. Azure AD authentication in a multi-cloud enterprise setup.
Received multiple SpotOn, Manager Appreciation, and Peer Appreciation awards during tenure.

.NET CoreAngularKafka / MSKAWS GlueSQSLambdaAPI Gateway WebSocketsDynamoDBECS / ECRSageMakerHangfireAzure ADOPC/Kepware

Software Engineer

Jan 2019 – Dec 2020

Full-stack delivery across enterprise applications: CRUD workflows, backend services, database changes, and frontend screens using .NET Core, Angular, SQL Server, Entity Framework.
Built real-time visualization features for operational/Andon screens in automobile and retail domains.
Projects: Promotion Planning Management (Retail), CDTS Kaizen (Automobile), CDTS Modernization.

.NET CoreAngularSQL ServerEntity FrameworkREST APIs

Campus Mind · Linde (Internship) Aug 2017 – Jan 2019

Built MyCar — an eCommerce car marketplace with JWT role-based auth, buyer/seller/admin flows using .NET Framework and Angular.
Won weekly hackathons during training period at Campus Mind.
At Linde R&D/IT, automated lengthy workflows using UiPath, PyAutoGUI, OpenCV to reduce human effort significantly.

03 · Projects

Selected builds &
experiments.

⭐ Platform · Flagship ● Recent

CSX Scheduler

In-house distributed job orchestration platform that replaced a legacy JAMS scheduler contract. Architected from scratch with AWS SQS queue-driven execution, Quartz.NET, .NET Background Services, Windows services, processor agents, ALB + Route53 endpoints, blue-green deployments, and GovCloud support. Reduced annual operating cost from $700k → $110k/yr. Full ownership from design to production.

C# / .NET 8Quartz.NETAWS SQSALBRoute53CloudFormationAurora PostgreSQL

🏭 Industrial IoT · Real-Time

Toyota Real-Time Factory Monitoring

End-to-end industrial IoT and real-time monitoring platform for Toyota manufacturing. On-prem machine KPIs via OPC/Kepware → Hangfire .NET ingestion → Kafka/MSK → AWS Glue (transform + PostgreSQL RDS) → SQS → Lambda → API Gateway WebSockets → targeted Andon dashboards. DynamoDB for connection routing. SageMaker Isolation Forest anomaly detection surfaced as live warnings. Backend APIs in .NET Core containerized via Docker → ECR → ECS, with Azure AD auth.

Kafka / MSKAWS GlueAPI Gateway WebSocketsDynamoDBSageMakerECS / ECRHangfireOPC/KepwareAzure AD

🤖 AI / Multi-agent Pipeline ● Recent

Project Harvest

Multi-agent document intelligence pipeline for generating structured HTML/PDF output at scale. Data Agent pulls form config (field definitions + HTML design template) and user records (name, Gov. ID, contact, FORM_ID). On unknown forms, Vision Agent runs DOCLING OCR to extract fields from the raw document. On known forms, Mapping Agent resolves fields via AI field matching or direct match. Validator Agent applies field formatting and validation rules. Publisher Agent pulls the appropriate base HTML template and renders the final HTML/PDF output.

Multi-agentDOCLING OCRVisionAI Field MatchingDocument IntelligenceHTML/PDF rendering

🧠 AI / LLM Fine-tuning ● Recent

CSOD-R1 & CSOD-IC

Designed and trained an internal MoE-style LLM from scratch — fine-tuned Qwen2.5-7B-Instruct with LoRA/PEFT on AWS G5.2xlarge (NVIDIA A10G), rank 128, 8-bit loading, ~$10/training run. Built CSOD-IC intent classifier on DistilBERT for routing. Converted to GGUF/Ollama, deployed via vLLM, and built the full backend + frontend integration. Load tested to 150 concurrent requests with stable p95.

Qwen2.5-7BLoRA / PEFTvLLMDistilBERTAWS G5GGUFOllama

🔌 AI / Developer Tools

MCP Servers

Built Model Context Protocol servers for CSX Scheduler and FeedStats applications — enabling AI agents to query and interact with internal enterprise systems through structured tool interfaces.

MCPTypeScriptEnterprise AI

🧪 AI / VS Code Extension

Pebbles

VS Code extension for AI-assisted unit test generation using internal LLM endpoints and LM Studio. Designed around selected source files, prompt execution, and generated test output.

TypeScriptVS Code APILM StudioRAG

🤖 Framework · .NET ● Recent

Squad.NET

A .NET-based agent orchestration framework inspired by CrewAI. Focuses on typed tasks, structured LLM output enforcement, JSON schema validation, and business-friendly orchestration vocabulary — built for .NET-native advantages.

C#Agent OrchestrationJSON Schema

View project →

📱 iOS · Product

BorderApp for iOS

Offline-first iOS photo utility for adding exhibition-style borders to images. Ranked Top #25 in the App Store Photos category. No login, no tracking, no subscriptions. Shipped through the full App Store release process.

View on App Store →

🔧 Infrastructure

Feed Tool Deployment Automation

Built a master orchestrator pipeline that accepts start/end stage, identifies applicable accounts/regions/servers, and parallelizes feed tool deployment across all environments. Integrated with JFrog Artifactory for artifact distribution.

JenkinsGroovyPowerShellArtifactory

04 · Skills

Technologies I
work with.

Languages

C# Python TypeScript JavaScript SQL PowerShell Groovy Rust (basic)

Backend & Distributed Systems

.NET / ASP.NET Core Quartz.NET Kafka / MSK Hangfire Background Services Windows Services SQS Workers C# Channels REST APIs WebSockets Microservices OPC/Kepware ingestion

Cloud & Infrastructure

AWS SQS CloudFormation ALB / Route53 AWS Glue Kafka / MSK EC2 / ASG Lambda S3 ECS / ECR API Gateway WebSockets DynamoDB Aurora PostgreSQL SageMaker IAM / KMS SSM / Secrets Manager GovCloud / FedRAMP

AI / ML & Agentic

LLM Fine-tuning (LoRA/PEFT) vLLM RAG MCP Qwen2.5-7B-Instruct DistilBERT Hugging Face Transformers PyTorch Isolation Forest SageMaker Multi-agent workflows OCR / Vision Adaptive learning / Q-learning GGUF / Ollama Scikit-learn LIME / SHAP AI red-teaming Quantization

Databases

PostgreSQL Aurora PostgreSQL DynamoDB SQL Server PostgreSQL RDS Entity Framework

DevOps & Tooling

Jenkins Docker GitHub JFrog Artifactory SonarQube XL Release Jira PowerShell automation

Frontend

Angular React TypeScript Swift / SwiftUI HTML / CSS

Testing

xUnit Moq NUnit Jasmine / Karma AI-assisted test gen

07 · Publications

Research &
academic work.

Thesis

Personalizing Employee Learning in the Tech Sector: Implementing AI-Based Procedural Content Generation in Corporate Training

BITS Pilani · IEEE · Zenodo March 2025 · M.Tech Dissertation

Develops a novel AI-driven framework using LLMs, Transformer-based models, and GPT variants to generate adaptive, personalized corporate training content on the fly. Combines supervised learning for knowledge classification with reinforcement learning agents that tune difficulty, complexity, and content sequencing based on real-time performance signals. Implements role-specific scenario generation across Bloom's taxonomy levels and a Q-learning-style adaptive loop tracking engagement and performance history.

↗ DOI · Zenodo ⬇ PDF

Preprint

When Valid Is Not Faithful: Verifier-Calibrated Search and Repair for Structured Generation

OSF · Open Science Framework DOI: 10.17605/OSF.IO/8TJMV · Python · TeX · 93% Python codebase

A research framework tackling faithful structured generation — where generated artifacts can be syntactically valid but semantically wrong. Instantiated on natural-language to PDDL translation (Planetarium blocksworld & gripper domains). Combines best-of-K candidate generation, a learned cross-encoder semantic verifier trained through iterative hard-negative mining and ranking-aligned retraining, and domain-aware one-step repair with non-oracle feedback. Key result: repair-augmented VCSR improved mean K=8 semantic equivalence from 0.4200 → 0.7720 on untouched held-out seeds. Post-freeze Claude-family benchmark (Haiku 4.5, Sonnet 4.5, Opus 4.6) showed VCSR repair consistently improved prompt-only K=8 generation across all tested models.

↗ DOI · OSF ↗ OSF Project ⌥ GitHub

Pritam Mondal.

Building systems
that scale and last.

Where I've built
things that matter.

Selected builds &
experiments.

Technologies I
work with.

Academic &
professional grounding.

Awards &
achievements.

Research &
academic work.

Let's build
something great.

Pritam Mondal.

Building systemsthat scale and last.

Where I've builtthings that matter.

Selected builds &experiments.

Technologies Iwork with.

Academic &professional grounding.

Awards &achievements.

Research &academic work.

Let's buildsomething great.

Building systems
that scale and last.

Where I've built
things that matter.

Selected builds &
experiments.

Technologies I
work with.

Academic &
professional grounding.

Awards &
achievements.

Research &
academic work.

Let's build
something great.