RoosCloset Labs engineers AWS-native intelligence to solve massive, overlooked inefficiencies. We help fashion enterprises reclaim $100B in lost returns, and we are transforming how 170 million Bengali citizens navigate civic life.
Fashion e-commerce bleeds revenue because it treats visual catalogs like flat spreadsheets. Meanwhile, millions of citizens face an invisible wall of bureaucracy, lacking a unified digital guide that actually speaks their language. Both are fixable with applied machine learning.
E-commerce search fails because product data is written for warehouses, not for human aesthetics. ATLAS eliminates this at the source.
The fashion industry loses $100B annually to returns. 47% are caused by sizing; 31% by photography mismatch. These aren't random events—they are predictable and preventable.
Navigating government processes requires insider knowledge. Bangladesh has 170M citizens who lack a centralized, accessible way to create official documents.
ATLAS takes a raw product image and breathes life into dead data. It returns 180+ structured attributes, a 512-dimensional style embedding, and consumer-ready copy—in under 60 seconds.
Merchant product data is built for inventory systems, but customers shop with aesthetic intent. ATLAS bridges this gap by looking at a garment and instantly extracting its semantic DNA—translating pixels into a language your search engine actually understands.
The result is transformative: Search that finds what customers mean, not just what they typed. Recommendations that grasp personal style. And critical return risk flags raised before a product ever goes live on your site.
The fashion industry is drowning in dashboards that offer no real solutions. MIRROR changes the paradigm using causal AI. We score every checkout in 150ms and isolate the exact reason a return will happen.
Prediction without causal explanation is useless. MIRROR is built on DoWhy causal inference—the same framework used in advanced academic research—trained directly on your returns data to identify true causal roots versus mere correlations.
When MIRROR tells your merchandising team that adding a specific size chart will reduce returns by 12%, it is providing a backdoor-adjusted causal estimate. It prescribes ranked interventions with measurable ROI, transforming insight into immediate action.
XGBoost scores return probability at checkout using 200 features. Sub-150ms P99 latency. Runs synchronously at the order API boundary.
DoWhy causal graph runs async on high-risk orders. Identifies true causal roots with backdoor adjustment for confounders.
Bedrock Claude generates ranked interventions with ROI estimates specific to your SKU. Surfaces in embedded QuickSight dashboards.
Navigating bureaucracy shouldn't require insider knowledge. BoroBhai is a centralized, AI-powered civic hub that democratizes access to state services for 170 million citizens.
BoroBhai is the single destination to get answers and take action. Ask a question in everyday Bengali or Banglish, and it instantly retrieves the correct workflow from a rigorously reviewed civic knowledge base. But it doesn't stop at advice—it actively creates the government forms, trade licenses, and legal agreements you need to move forward.
By shifting the burden of bureaucracy from the citizen to the machine, BoroBhai ensures no one is left behind by the digital divide. It is enterprise-grade infrastructure deployed for public good.
Speak naturally. BoroBhai natively understands formal Bengali, conversational slang, and mixed-script Banglish.
Stop hunting for templates. Instantly generate perfectly formatted trade licenses, leave applications, and agreements.
Built for everyone. High-accuracy Bengali speech recognition allows citizens to navigate using only their voice.
Merge, compress, and convert your official PDFs directly within the chat window to meet government portal requirements.
We don't build API wrappers. RoosCloset Labs engineers 100% AWS-native infrastructure deployed via CDK. Our platforms feature zero-trust multi-tenant isolation, real-time event streaming, and automated continuous retraining pipelines out of the box.
Multi-tenant isolation isn't a convention; it is cryptographically enforced. We utilize tenant-prefixed S3 paths, partition-keyed DynamoDB architectures, per-tenant OpenSearch indices, and Cognito with custom identity claims to ensure absolute data sovereignty for every client.
A track record of engineering mathematically rigorous, production-grade AI systems. Rather than building theoretical models, Najmun specializes in dismantling complex industry bottlenecks and reconstructing them into scalable, cost-optimized AWS infrastructure.
Previous deployments include physics-informed SegFormer architectures for flood detection, Gaussian Process Regression for epidemiological forecasting (9.8% MAPE), and SMS-based agricultural advisory systems scaling to rural populations at $0.001 per inference. Every system is in production. None are demos.