I design and ship LLM-powered systems, document intelligence pipelines, and real-time AI workflows — turning ambiguous, high-friction problems into production software that scales.
Currently Software Engineer at Randomwalk AI — designing LLM backends for document intelligence, async job orchestration and enterprise automation.
Randomwalk AI · In production
End-to-end platform for document classification, extraction and validation using LLMs — bulk processing with async job orchestration, turning hours of manual review into minutes.
My work sits at the intersection of LLMs, backend engineering and pragmatic system design. I care deeply about latency, reliability and cost — and about making AI systems feel less like research experiments and more like infrastructure you can rely on.
Previously I built frontend experiences at docu3C, competed in national hackathons, and earned an NVIDIA Deep Learning Institute certification in AI & Accelerated Computing.
Designing and deploying LLM-powered document processing systems for financial data — bank statements, transactions, reconciliation. Built async bulk-job pipelines, AI-driven nudging systems, and optimized workflows for cost, latency and reliability (retry, fallback, guardrails).
Built backend microservices and REST APIs powering early AI applications. Automated internal workflows and shipped production features in tight agile cycles — the runway into a full-time role.
Developed responsive React interfaces and collaborated with design to raise usability and accessibility — where the obsession with polish and detail started.
End-to-end platform for document classification, extraction and validation using LLMs. Backend architected for bulk processing with async job orchestration — turning hours of manual review into minutes.
An organization-wide engagement platform integrating Microsoft Graph APIs with an AI-generated nudging engine — surfacing the right message, to the right person, at the right moment.
A live pipeline ingesting RTSP streams and running activity detection on the edge — with alerting and periodic AI-generated summaries so humans only look when it matters.
Lightweight multi-user streaming system with frame-accurate synchronized playback — a study in real-time state, low-latency messaging, and keeping simple things simple.
