Oblivious
May 2023 – Present
Software Engineer
Bangalore, Karnataka
- Architected a multi-provider LLM gateway/proxy (Anthropic, OpenAI, Gemini) fronting AI coding tools (Claude Code, Cursor, Copilot, Codex), serving 15K requests/day across 5 engineering teams
- Built prompt/tool-call tracing that links every Jira ticket, PR, commit, and diff to the underlying AI prompts, skills, and MCP tools used — giving leadership full auditability of AI-generated code
- Designed a unified request/response schema normalizing Anthropic, OpenAI, and Gemini APIs, with model routing, per-team rate limiting, and in-flight request intervention for policy enforcement
- Built PII/secrets redaction that detects and tokenizes sensitive data (API keys, passwords, credentials) in prompts before they reach external LLM providers, preventing data leakage from AI coding tools
- Shipped a usage analytics pipeline surfacing per-team, per-repo AI adoption metrics, used by engineering leadership to quantify AI ROI and prioritize tool rollout
- Created OBLV Deploy supporting AWS Nitro Enclaves, Azure Confidential VMs (AMD SEV-SNP attestation), and NVIDIA H100 confidential GPUs for multi-cloud confidential ML workloads
- Built a Layer 4 Rust proxy (TCP/UDP over vsock), improving enclave service throughput 30% and enabling secure external connectivity