Aryan Marwah

Software Engineer · Peoria, IL

• Lead a team of 5 engineers building an end-to-end on-device RAG pipeline on the SiMa.ai Modalix DevKit using the LLiMa framework — ingesting, chunking, and embedding ~5,000 pages of agricultural equipment manuals into a fully offline vector retrieval system with zero cloud dependency.
• Built Hey Sim, a multimodal voice and vision assistant deployed entirely on the Modalix DevKit, letting field technicians ask natural-language questions about equipment and receive real-time visual guidance on which controls to interact with — no internet connection required.
• Engineered the full application stack from document chunking strategy through the Flask web app, with on-device speech-to-text, vision inference, and LLM response all running inside a sub-10W power envelope.

Software Engineer · Germantown, MD

• Built a telemetry ingestion pipeline landing ~600GB/day of satellite data in AWS S3, processed with Glue and Lambda into partitioned datasets that let ops and analytics teams run queries and build dashboards.
• Built an internal LLM-driven support agent that reads error codes, looks up fixes, and safely triggers actions like controlled server restarts — automatically handling about 60% of recurring tickets.
• Optimized a distributed anomaly-detection pipeline processing 100k+ telemetry data points by parallelizing across threads and containerized ECS workloads, cutting runtime from 3 hours to ~40 minutes.
• Shipped a domain-aware knowledge assistant using BGE-1.5, Qdrant, and Gemini with LoRA fine-tuning, giving engineers instant semantic lookup and context-rich answers.
• Built a Python anomaly detector for terminal telemetry using Isolation Forest, k-NN, and HBOS, tuning features and alert thresholds with domain experts and writing daily risk reports to S3/NAS.

Machine Learning Intern · Germantown, MD

• Built a two-tier satellite interference detector (CNNs plus anomaly detection) with a custom pseudo-data generator emulating ADC visualizer output to enrich training — improving accuracy by 30% and surfacing previously unseen interference patterns alongside a 4-engineer team.

Software Engineer Intern · Peoria, IL

• Engineered a farming simulator in C and WPF, integrating physical hardware to mimic real-life farming scenarios and driving a 70% increase in agricultural tool comprehension.

B.S. Computer Science

Coursework: Distributed Systems, Machine Learning, Computer and Network Security, Parallel Computing