Apr 2025 – Present · 1 yr 1 mo

• 🤖 Codelassian: AI Coding Agent
• Full-stack engineer building AI coding agents and AI products (UX + infra). Had the privilege to work with some exceptional engineers while shipping this.
• DAU: 1K+ engineers
• MAU: 3K+ engineers
• ~95% Reusable Code cross-platform (VsCode, JetBrains)
• Built the AI-native editor + chat UX end-to-end: persistent + conversation editors, attachments, @ mentions, '/‘ commands, image paste/upload, files drag-and-drop, and keyboard-first interaction.
• Improved AI Agent’s answer quality by building structured context → prompts (incl. XML-formatted context blocks + transformations) so the agent gets the right context and returns most relevant actions/answers.
• Shipped Context providers for AI Agent: ‘@Files & Folders', '@Commit', ‘@Branch’, ‘@Rules’, ‘@Docs' (internal documentation semantic index), ‘@Linter Errors’, ‘@Recent Changes', ‘@Past Chats'
• Extended AI Agent’s capabilities with MCP (infra + UI): connect to MCP servers, discover tools, manage per-server connection state, live refresh, and safely let users interact with third-party systems and take actions through the agent.
• Improved agent’s file-editing correctness by tightening the feedback loop: non-intrusive VS Code diagnostic collection + a context selector for @Linter Errors so the agent can fix what the IDE is actually complaining about.
• Contributed to delivering a cross-IDE platform + feature-driven architecture (VS Code + JetBrains) with ~95% code reuse, minimizing duplication and increasing shipping velocity for the team.
• Built Extension + Agent Settings end-to-end: infra, persistence, real-time updates, and UI/UX for configuration (including permissions/approvals and extensibility surfaces).

Jan 2024 – Apr 2025 · 1 yr 3 mos

• 🧪 Predictive Test Selection
• Contributed in the MLOps space: we shipped ML models to predict which tests (Unit, Integration, VR tests) should run per code change in local and CI/CD pipelines using changed files and historical failure signals
• Reduced the number of tests executed in build pipelines, and achieved faster build times, tighter feedback loop, improved developer productivity and decreased infra costs.
• Drove ~99% improvement in p95 prediction API Latency (60s -> 500ms) by implementing 2 out of 3 key upgrades - Tecton Feature Store Integration and Redis Caching (alongside Flask -> FastAPI migration)
• Integrated Tecton as a feature store to make feature computation consistent across training + serving, enable low-latency online feature retrieval, and eliminate drift between offline pipelines and real-time predictions.
• Contributed to MLOps Infrastructure automation across the lifecycle: data ingestion, validation, labelling/splitting, training/retraining, testing, deployment and serving