Sachin Aggarwal

About Sachin Aggarwal

I build production-grade AI systems that actually work in the real world.Over the past few years, I’ve been focused on Applied AI—designing and scaling LLM-powered systems for complex, unstructured workflows like document processing. My work sits at the intersection of backend systems, AI, UI and platform engineering.Currently leading development of an Intelligent Document Processing (IDP) platform at Avalara, processing ~150K documents/week through end-to-end pipelines involving OCR, classification, extraction, and summarization.Key areas I work on:• LLM & Agentic Systems: multi-stage workflows with confidence scoring, citations, human-in-the-loop review, and evaluation loops • AI Platform Engineering: building reliable pipelines using Temporal, queues, and distributed processing • Applied AI Optimization: improving accuracy (60% → 92%) via automated prompt tuning, evaluation systems (GEPA, TextGrad), and feedback loops • Scalability & Reliability: handling large documents with parallel processing, retries, observability (Prometheus, Grafana), and cost optimization • I also bring strong product and user experience intuition—having built frontend systems (React/TypeScript) and actively shaping user journeys, review workflows, and human-in-the-loop interfaces for AI systems.Tech stack: Python, FastAPI, LLMs, RAG, Temporal, Redis, AWS, Kubernetes, React.js, Vue.jsI’m particularly interested in:• Applied AI / GenAI systems • Agentic workflows & orchestration • AI infrastructure & platform engineering

Experience

1. Technical Lead

   Avalara

   Jul 2024 — Present

   Leading the development of a scalable Intelligent Document Processing (IDP) platform, supporting end-to-end workflows across OCR, extraction, classification, summarization, etc.• Architected Temporal-based distributed workflows with chunking and parallel processing for large documents, enabling reliable OCR + LLM pipelines at scale (~150K docs/week).• Built a production-grade extraction agentic system for documents - confidence scoring, HITL review, citations, redactions, and evaluation pipelines for reliable and explainable LLM workflows.• Improved extraction accuracy from ~60%->~92% via closed-loop system: auto prompt tuning driven by evaluation pipelines, ground truth, and user feedback (GEPA, TextGrad).• Resolved memory retention and spike issues arising from large document/image processing, reducing memory usage by around 40% and stabilizing production workloads.• Implemented observability (Prometheus, Grafana, tracing) to improve system reliability, debuggability, and platform stability.

Education

• Maharaja Agrasen Institute Of Technology, Delhi

  Engineer’s Degree, Information Technology

Skills

• Java
• Android Development
• Python
• Ubuntu
• Android Studio
• Data Structures
• Css
• Web Scraping
• Management
• Linux
• Html
• Eclipse
• Mobile Application Development
• C and Graphics
• Git
• Wireshark
• Hadoop
• Mysql
• Github
• Netbeans
• C++