Hi, I am Aditya 
Senior Software Engineer
I am a Full-Stack Software Engineer (expertise in backend) with over around 5.5 years of professional experience specializing in high-throughput backend systems, modern DevOps practices, and autonomous agentic AI workflows.
I design and scale systems serving over 1,000,000+ users, leading critical infrastructural upgrades (Kafka, JDK, and Spring migration) and building secure, enterprise-grade AI automation pipelines.
Github
LinkedIn
Kaggle
Resume
IELTS (7.0)Beiong E-Commerce
Looking for my latest work? Explore Beiong, a modern, client-facing e-commerce platform built with Java 21, Spring Cloud, AWS, and Next.js.
Skills & Experience
Professional Experience
Therap Services LLC 🔗
Around 5.5 years of Software Engineering experience, specialized in developing and managing complex enterprise systems. Led core projects, delivering high-quality, high-throughput systems. Strong expertise in code review, cross-team collaboration, mentoring, and automation in an agile environment using Scrum practices. Focused on building highly secure, reliable, scalable, and robust systems compliant with strict US federal guidelines.
Sr. Software Engineer
April 2026 - Present- Architected an end-to-end RAG pipeline using Python and LangChain, engineering autonomous Agents with custom Tool-calling capabilities—including real-time user access and permission checks—to enable secure querying across a 10,000+ document enterprise knowledge base.
- Migrated core legacy messaging infrastructure from ActiveMQ to Apache Kafka, introducing a phased migration strategy with a custom synchronization bridge ensuring zero message loss (outbox pattern), resulting in a 70% throughput gain and 40% latency reduction.
- Managed the JDK 8 to JDK 21 upgrade of a large legacy codebase, planning and managing in-app and service dependencies, leveraging OpenRewrite and custom recipes for efficient execution.
Software Engineer II
April 2024 - March 2026- Worked in a team of 10+ members to maintain core system functionalities, including login, EVV (Electronic Visit Validation), Scheduling (Google Calendar-like), and Secure Messaging (Gmail-like secure feature).
- Architected and developed Webhooks for Therap's aggregator application, enabling efficient system-to-system communication and data integration.
- Re-engineered high-volume cron jobs and data access patterns using Oracle Coherence, advanced query optimization, and intelligent indexing, reducing database QPS and achieving a 20% reduction in latency.
- Load-tested API endpoints with Gatling and JMeter, identifying bottlenecks in the legacy UUID generator, and replaced it with Twitter Snowflake UUID, achieving 5x insert/update operations per second.
- Dockerized multiple legacy services, including the authentication server, and developed a single-click deployment tool, cutting deployment time by 50% and reducing developer hardware requirements by 33%.
- Designed data-integration pipelines and automated manual workflows for SFTP and REST APIs, enabling Therap to secure a multi-million dollar Kentucky state project.
- Rearchitected data import pipelines, enabling 20x faster processing for large imports (150K+ rows) using queue and batching techniques.
- Mentored junior developers and led technical sharing sessions.
Software Engineer
April 2022 - March 2024- Integrated Salesforce and MaestroQA APIs, streamlining customer support workflows and reducing manual intervention by up to 80%.
- Contributed to implementing an in-house authentication module, RBAC, and ABAC using Spring Security featuring MFA, OAuth2, JWT, and rate-limiting, reducing operational licensing costs to zero.
- Delivered state-specific HIPAA-compliant solutions for Kentucky and Tennessee, implementing rollup-dashboards and meeting the federal 21st Century Cures Act mandate, securing multi-million dollar state contracts.
- Developed data-feed dashboards processing over 1 million records per day using Spring Batch, JDBC, and Redis with asynchronous multithreading for real-time monitoring.
- Architected and developed high-performant data integration pipeline with multiple Giant health care providers like HHA, RHA, Sandata, and Federal States (TN, KY) leveraging REST APIs, and adhering to OAS 3.0.
- Involved in Developer and QA-side On-premise server deployments using WebLogic, Tomcat, Nginx Proxies and Reverse Proxies, and Kafka Zookeeper ensuring high performance and stability.
Associate Software Engineer
March 2021 - March 2022- Developed and maintained a large-scale customer-facing B2B SaaS application with 40,000+ Daily CCU (Concurrent Users) serving over 1,000,000+ individuals.
- Optimized on-demand loading of resources and integrated CDNs, reducing network bandwidth by 20% and improving page load times.
- Developed an in-house message translation tracker for localization of the application, fully eliminating manual intervention.
Technical Expertise
Programming Languages
Frameworks & Libraries
Cloud, DevOps & Infrastructure
Databases & Caching
Data Streaming & Middleware
Machine Learning & AI
Agentic AI Systems
- Autonomous Development Workflows (/plan, /design, /implement, /test, /review)
- AI Agent Persona Orchestration (Product Manager, Architect, Senior Dev, SDET, Security Auditor)
- Context-aware planning, HITL (Human-in-the-Loop) guardrails, validation checklists
Research & Publications
Study of Spanning tree with Maximum number of leaves
Spanning Tree of a graph is a tree where all nodes are connected. A graph might have multiple spanning trees. A maximum leaves spanning tree (MLST) is a tree whose number of leaves is maximum. Finding MLST of a graph is a NP-hard problem. It is also APX-hard, which means its approximation ratio is bounded by a constant. By finding the MLST of a network, we can find the backbone of the network, which can be used to reduce the cost of the network. In this work, we tried to find an approximate algorithm for the MLST of a graph. We provide a heuristic algorithm following a bottom-up approach.
Chakma Handwritten Characters recognition and Multipurpose Dataset
The Chakma language is an endangered language spoken by the Chakma people. Not so many resources are available in the Chakma language. Here we introduce a novel multilabel dataset for the Chakma language. The dataset contains 111,382 handwritten character images. We also introduce a novel data collection method and provide a baseline model, comparing with state-of-the-art models. We provide a novel pipeline for handwritten character recognition. First we convert handwritten characters to font type images using GAN, then we use a Visual Transformer based model for classification.
Multilingual machine translation for Chakma
In this work we introduce a novel bilingual and monolingual corpus for the Chakma language. The dataset consists of 13,000 sentences, collected from various sources including online crowd sourcing. We then experiment with Neural Machine Translation (NMT) and Statistical Machine Translation (SMT).
Android malware detection
In this work we curated a dataset from various sources. Then reverse engineered the APKs to form source code analyzing using APKtool for decompiling. We made a code parser and extracted import, method invocation, and parameter-based features from the source code. Afterwards, recursive feature elimination and PCA was used for feature reduction. Then we ran tree-based models for classification of benign and malicious apps. We ran feature importance on the collected dataset and found that, method invocation based features are the most important features.
Selected Projects
Agentic Coding and Agentic SDLC Automation
Multi-Agent Pipelines
Orchestrates specialized, context-aware AI agents across the software development lifecycle using secure command sets:
/plan PM persona break down epic roadmap requirements/design Architect maps out interfaces and system diagrams/implement Dev designs clean, strongly-typed component blocks/test SDET codes extensive automated unit/integration suites/review Auditor audits static code for safety and optimizationRepository Context Maps
Enables reliable autonomous tasks by locking down state dependencies, schemas, and lint standards directly in the codebase:
- dependency.md: Explicitly constraints dependency bloatedness
- architecture.md: Outlines state trees and network client schemas
- standards.md: Strict lint standards and quality guardrails
Human-in-the-Loop Safeguards
Enforces architectural alignment through active, manual sign-off checks combined with strict verification checkers:
- Verification Gates: Pause execution after plan/design phases
- Dry-Checks: Automated checklist verification across DRY compliance
Enterprise RAG & Tool-Calling
Designed and deployed secure, production-grade agentic architectures with advanced capabilities:
- End-to-End RAG Pipelines: Engineered semantic search using Python, LangChain, and Vector Databases over a 10,000+ document enterprise repository.
- Autonomous Tool-Calling: Built specialized Agents equipped with custom APIs, executing dynamic real-time permission and access-control checks.
Featured & Personal Systems
Beiong
Featured Full-StackBeiong is a modern, client-facing e-commerce platform designed for global scalability. It is built using a highly scalable microservices architecture on the backend and a mobile-first, TanStack Query-powered Next.js interface on the frontend. The project utilizes containerization and advanced AWS services for secure, high-performance commerce management.
Technologies
Key Capabilities
- Catalog Control: Full CRUD operations for Products & Categories with type-safe schema validation
- Promotion Management: Admin interface for Featured Products & Featured Categories storefront highlights
- Order Oversight: Secure order queues to monitor, filter, and track fulfillment status
- Advanced Attachment Service: Custom S3 multi-file upload flows via secure, time-limited presigned URLs
Online Shop
Featured Full-StackAn e-commerce online shop website. Users can open their shop and sell products online by posting them. After site admin validates and accepts the products, they will be visible to customers. Customer can place order for the products.
Technologies
Key Capabilities
- Secure Admin Control: Dashboard for administrators to login, add, update, and discontinue products
- Fulfillment Pipeline: Admin order management queues to confirm or reject pending customer requests
- Product Catalog: Searchable public index allowing customers to discover and filter products by name
- Shopping Cart System: Interactive client-side cart allowing users to add products and proceed to checkout
Attention (Google Chrome Extension)
Featured Full-StackDeveloped a Chrome extension to enhance focus by blocking distracting websites, hiding YouTube shorts and Facebook reels, and bookmarking YouTube timestamps. Built using Vanilla JavaScript, Manifest V3, HTML, and CSS, with planned enhancements for customization and LLM-based content summarization.
Technologies
Key Capabilities
- Timestamp Bookmarking: Custom bookmarking module to save specific playback moments in YouTube videos
- Reel Filtering: Content scripts to dynamically block and hide Facebook Reels from the homepage newsfeed
- Shorts Removal: DOM injection styles to filter and hide YouTube Shorts from homepages and suggestions
- Website Blocking: Customizable distraction blocker to restrict access to user-specified domains
Academic & Structural Systems
Structured Programming Language: DX-Ball
Academic SystemThe 90's classic DX-Ball game re-created using C++, OpenGL, and iGraphics library. The iGraphics is a graphics library for C++ made by a final year student. This game has 5 different levels, lives system, and scoring system.
OOP: Blood Bank
Academic SystemUsed JavaFx to build a blood donation desktop application. Users and donors can register in the system. Upon users request for blood, donors in the corresponding location will get notified.
Software Engineering: Travel Mate
Academic SystemTravel Mate is a web application designed for managing tour agencies. It caters to two primary user groups: business owners and tourists. Business owners can register their businesses and list services such as hotel bookings and vehicle rentals. Tourists can sign up to explore and book these services. The platform tracks the availability of rooms and vehicles, updating their status in real-time to reflect current occupancy.
Neural Networks: LCU-Net
Academic SystemU-Net is an encoder-decoder based CNN architecture primarily designed for EMs (Environmental micro-organisms) segmentation. LCU-Net is a version of U-Net using much lower memory and computational power. I implemented the paper 'LCU-Net: A novel low-cost U-Net for environmental microorganism image segmentation' from scratch in python.
Education
Master of Cyber Security
University of Newcastle Australia
M.Sc. in Computer Science and Engineering
Bangladesh University of Engineering and Technology (BUET)
Key Focus Courses
Thesis Topic
Chakma Handwritten Characters Recognition and Multipurpose Dataset
OngoingB.Sc. in Computer Science and Engineering
Bangladesh University of Engineering and Technology (BUET)
Key Focus Courses
Thesis Topic
Study of Spanning Tree with Maximum Number of Leaves
CompletedCertifications & Achievements
Olympiad & Competitive Highlights
- ★ SIIM-ISIC Melanoma Classification (Kaggle)
- ★ PetFinder.my - Cuteness Predictor (Kaggle)
- ★ Bengali.AI - Grapheme Classification (Kaggle)
- ★ Dhaka AI - Traffic Detection Challenge
- ★ IUT National ICT Fest Hackathon 2017
BDMO (Bangladesh Mathematics Olympiad)
Perfect Attendance Certificate
Awarded Perfect Attendance Certificate at Notre Dame College, Dhaka.