I build AI products that have to work in the real world.

Right now I'm focused on a mix of applied AI, product architecture, and technical leadership across privacy-sensitive, regulated, and high-stakes environments. Most of the work I care about sits at the intersection of product usefulness, model reliability, and infrastructure decisions that won't become a liability later.

• Building privacy-first AI products with a strong bias toward practical use, clean UX, and sustainable architecture
• Designing retrieval systems that are more deterministic, better scoped, and less prone to hallucination
• Shipping multimodal workflows that combine text, documents, visuals, and structured data
• Exploring sovereign and EU-hosted AI stacks for teams that care about compliance, residency, and vendor risk
• Turning messy operational problems into decision systems, simulations, and tools people can actually use day to day
• Growing small software products from idea to usable product without overengineering the first version

These are the kinds of problems I'm deep in at the moment:

• Knowledge retrieval tools with hard guardrails, filtered context, and evaluation loops
• AI-assisted systems for document-heavy and research-heavy workflows
• Computer vision workflows for messy real-world environments
• Decision-support systems that use causal thinking instead of just prediction
• Lightweight SaaS products with strong product sense and careful cost control
• Architecture reviews for early-stage products figuring out what to build in-house vs what to outsource to vendors

• RAG systems
• multimodal AI applications
• privacy-aware AI workflows
• internal tools that save teams real time
• MVPs that can grow into actual products
• evaluation layers for LLM-based systems
• infrastructure setups that stay understandable six months later

Python TypeScript SQL PostgreSQL Docker PyTorch scikit-learn BigQuery Airflow

Also spending a lot of time around:

RAG multimodal systems causal inference computer vision Mistral Gemini AWS GCP OVHcloud Scaleway

• start from the actual business bottleneck
• reduce ambiguity early
• build small, verify fast
• keep AI systems observable
• avoid magic where reliability matters
• write things down so future decisions are easier

I’m especially interested in:

• AI products with real operational value
• European AI infrastructure and sovereignty
• product strategy for small technical teams
• the gap between demo AI and deployable AI

• Website: arubhardwaj.eu
• LinkedIn: linkedin.com/in/arub
• GitHub: github.com/arubhardwaj