Production software around SpaceXAl's model stack: APls, internal tools, and the pipelines that sit between data, models, and product.

Built services and evaluation loops used in training and product, ship APls, and keep data moving through the systems that serve Grok.
Part of end-to-end LLM evaluation and model training for software engineering applications, developing coding benchmarks, implementing automated evaluation pipelines, analyzing model behavior, and driving iterative improvements that increased model quality and reliability in production.

Large-scale evaluation pipelines for synthetic agent trajectories.

Engineered batch simulation, parallel worker orchestration, and fault-tolerant metric aggregation with Python, asyncio, and Redis task queues over 10K+ runs. Pandas and SQL analytics on top. Built automated evals for reward efficiency, progression stability, and economy balance, improving iteration speed for evaluation and tuning by 40%.

Built SDKs, APIs, and scalable systems powering cloud automation, data ingestion, and multi-tenant platforms.

Re-architected an i frame-based SDK, enabling seamless cross-domain integration and high adoption across clients. Developed performant APIs with for bulk ingestion, async queues, and third-party system integrations — supporting millions of records daily. Migrated 100+ legacy React components, eliminated 10+ memory leaks, and optimized frontend workflows by 10x with caching and lifecycle tuning. Contributed to micro-frontend migration, implemented RBAC search endpoints, and achieved 90%+ test coverage across critical services.

Led platform architecture and multi-tenant backend systems for an EdTech SaaS serving 25+ institutions.

Designed JWT-based auth with Spring Security, supporting both RBAC and ABAC for dynamic, tenant-aware access controls. Engineered PostgreSQL isolated-schema multi-tenancy with Liquibase migrations for safe onboarding and upgrades. Built resilient Spring Batch pipelines for CRM data migration from legacy systems, halving onboarding time. Developed CI/CD pipelines via GitHub Actions with build verification and zero-downtime releases, improving release velocity 3x. Tuned backend performance using JPA optimizations and Caffeine caching, reducing average API latency by 45%.

Automation-focused backend engineering for supply chain systems with cross-platform data integration.

Built an AI-driven document intelligence pipeline using Python, LLMs, and vector databases to extract key data from regulatory PDFs, invoices, and shipment logs, cutting manual reconciliation by 50% and improving planning accuracy. Developed a RAG system over 1K+ internal documents and structured data to deliver more relevant, domain-aware responses, improving evaluation scores by 35%, and created reliable ETL pipelines with Pandas and NumPy to validate schemas, clean high-volume unstructured data, and prepare type-safe datasets for downstream ML models.