Experience

Remote, India · Current

• — Building BoltzLabs Cloud, sandbox infrastructure that hands developers and AI agents an isolated machine through a single API: on-demand sandboxes that boot in about a second, persistent workspaces that release compute on pause and keep their files for agent sessions spanning days, and environments shaped for reinforcement-learning rollouts.
• — Shipped the platform end to end: Python and TypeScript SDKs, REST API, a cross-platform boltz CLI for macOS and Linux on Intel and ARM, authentication, dashboard, documentation, usage metering, SSH/SCP access and local port forwarding.
• — Built custom RL sandboxes on reserved capacity with sub-5ms step latency and caller-supplied images, alongside a Judge0-style execution API that keeps NumPy, SciPy and pandas resident so a run never waits on an install. GPU-attached sandboxes are next, behind the same API.
• — Designed the orchestration layer for fast provisioning and high density: warm reusable environments, automated allocation, resource limits, process execution, environment reset and clean respawn for repeated agent and RL workloads, metered per active second.

Remote, India · Current

• — Built a developer education platform where engineers implement production systems from scratch, including Redis, compilers, GPT/LLMs, databases, RAG systems, PyTorch internals and AlphaFold2.
• — Designed a trigger-based CI and evaluation workflow: a Git submission kicks off isolated test execution, per-challenge grading and structured feedback automatically.
• — Built a cross-platform desktop application, web product and custom CLI with integrated code editing, challenge browsing, test output and progress tracking.

Remote, India · Current

• — Built and operate a government-exam preparation platform combining mock tests, quizzes, current affairs, recruitment updates, results, admit cards, answer keys and AI-powered preparation tools.
• — Built a multi-model AI learning system with four specialised modes, routing tasks across different LLM workflows for UPSC question answering, answer writing, evaluation and structured exam-focused assistance.
• — Designed a LangGraph-based multi-agent evaluation pipeline that retrieves relevant context, analyses descriptive answers and generates structured feedback across relevance, content, reasoning, structure and presentation.
• — Engineered automated pipelines to collect, deduplicate, classify and publish exam updates and current affairs, with Redis-backed caching supporting high-frequency feeds and interactive practice.

Remote

• — Built production backend services and REST APIs in TypeScript, Express and MongoDB for a live assessment platform, supporting database-backed product features and application workflows.
• — Implemented AI workflow components and backend integrations with LangChain, connecting application data and APIs to LLM-powered product functionality.