Divyanshu Prasad

• •Contributed to an adaptive learning system using AI and human feedback loops, improving code review comment acceptance by ~18.5% (up to 28% for top customers).
• •Evaluated LLMs across tool-calling accuracy, quality, output variance, and cost, reducing per-PR inference spend by ~65%.
• •Re-architected enterprise custom guidelines from prompt-based execution to agentic workflows, integrating repository-level business rules into production AI workflows serving hundreds of pull requests weekly with ~29% suggestion acceptance.
• •Built Python based evaluation framework for production AI agents, supporting release validation, regression testing, and canary rollouts with continuous precision and recall monitoring.
• •Built a distributed Kafka pipeline in Go for asynchronous ingestion of historical pull requests, handling Git platform API rate limits to enable scalable AI workflow pre-tuning for enterprise customers.
• •Designed and implemented a vendor-agnostic file storage abstraction layer supporting NFS/EFS for on-premise AI agent deployments, eliminating cloud storage dependencies for enterprise customers.

—python · go · fast-api · generative-ai · prompt-engineering · mysql

• •Implemented Scala language support in the context builder module using Go and tree-sitter.
• •Led the YAML-to-JSON prompt migration across code analysis phases, standardizing output schemas for future workflows.
• •Developed Python automation to generate evaluation PRs from OSS datasets, enabling systematic code review quality assessment.
• •Led end-to-end evaluation of the code review agent across OSS and internal pull requests, analyzing precision, recall, context retrieval, and prompt strategies to identify quality bottlenecks.
• •Benchmarked competing code review agents and produced evaluation reports that informed product and engineering strategy.

—python · go · generative-ai · prompt-engineering

• •Improved landing page performance (40+ → 80+) via lazy loading, optimized media, and efficient animation handling.
• •Added multilingual support (EN, FR, DE, ES), expanding global accessibility and reach.
• •Built real-time dashboards with REST API integrations, optimized state management, and pagination across projects.
• •Developed CMS-driven pages and Sanity studios with dynamic SEO (OG tags, metadata), streamlining content workflows.
• •Built and integrated email templates for newsletters and onboarding, improving user engagement.

—javascript · svelte/sveltekit · tailwind-css · sanity-cms

• •Led a team of 9 contributors, managing tasks, reviewing code, and resolving blockers to deliver ahead of expectations.
• •Authored and presented project whitepaper to senior engineers, product team, and CEO.
• •Designed user flows and low-fidelity wireframes, enabling smooth handoff to product and design teams.
• •Built responsive frontend (Svelte, Tailwind) with Firebase integration; recognized among top 400+ contributors.

—javascript · svelte/sveltekit · tailwind-css · firebase · figma