Ashwin Tyagi

Architected a two-pass agentic pipeline that extracted and categorized product data from unstructured spec sheets, and created a filterable product schema: reduced manual data entry and improved catalog search ability for construction material sellers. Designed and validated the SKU attribute extraction schema by building evaluation pipelines to measure LLM output accuracy and iterating on prompts to improve the consistency of the structured data extraction. On the seller-facing chatbot, reduced AI inference costs by 25% per query by building a tiered intent-routing system that uses an SLM to classify a user's request and direct the request to its appropriate pipeline.

AI Software Engineer (Contract) - Construex - Remote

(2025-10 - 2026-05)

• Architected a two-pass agentic pipeline that extracted and categorized product data from unstructured spec sheets, and created a filterable product schema: reduced manual data entry and improved catalog search ability for construction material sellers.
• Designed and validated the SKU attribute extraction schema by building evaluation pipelines to measure LLM output accuracy and iterating on prompts to improve the consistency of the structured data extraction.
• On the seller-facing chatbot, reduced AI inference costs by 25% per query by building a tiered intent-routing system that uses an SLM to classify a user's request and direct the request to its appropriate pipeline.

Co-Founder & Full Stack Engineer - Abstrac - El Dorado Hills, CA

(2025-09 - 2026-02)

• Led full-stack development of a cross-platform social media app build on Expo (React Native) and Firebase, owning the architecture of real-time feeds, media workflows, and cloud services end-to-end.
• Designed and optimized backend infrastructure such as Cloud Functions APIs, Firestore schemas, composite indexes, and security rules; further implementing caching, batched reads, and CI/CD to reduce latency and control cloud costs.
• Built a client-side caching layer in Expo to reduce redundant Firebase reads and improve perceived load times across media-heavy screens, balancing UX responsiveness with cloud fetch overhead.

Software Engineer (Contract) - BrickRed Systems - Seattle, WA

(2025-04 - 2025-07)

• Optimized cloud-scale document AI pipeline by building an XGBoost + Optuna training workflow to filter non-informative real estate pages, cutting Azure OpenAI token costs while improving reliability of downstream LLM predictions.
• Designed and implemented a CI/CD pipeline in Azure DevOps to automate build, test, and deployment workflows for a multi-service AI document processing platform.

Bachelor of Engineering - Computer Engineering - University of California, San Diego (2020-09 - 2024-06)