Shravan Vadeghar

Engineered the core platform that lets developers rent, configure, and boot dedicated high-memory servers in seconds. Developers could jump straight into fully pre-configured build environments from their terminal or browser with zero manual setup.

Rewrote key parts of Google's upstream Ninja build system in C++ to traverse massive 180,000-node dependency graphs much faster, integrated with a tier-based Redis-compatible SLA engine enforcing dynamic multi-tenant concurrency quotas (12 to 96 vCPUs). This cut full Android 14 and 15 clean build times from 4+ hours down to under 40 minutes for ROM developers.

Architected our own complete Remote Execution API (REAPI v2) gRPC caching proxy and build engine from scratch in Go and Rust. Instead of depending on expensive third-party cloud providers, we built local ByteStream caching with RocksDB and Sled, federated a 3-node stateful cluster mesh, and validated Merkle tree cache integrity. This eliminated vendor lock-in, delivered sub-2ms cache lookups, and saved thousands in monthly cloud costs.

Engineered an automated AOSP build monitoring subsystem that detects Android compilation instantly, tracks build progress in real-time, and sends automated Telegram alerts with direct artifact download links the moment a build finishes.

Created an automated artifact pipeline that automatically catches interrupted uploads of large ROM zip files, verifying file freshness and resuming right where they dropped so developers never lose a 2-hour build.

Built custom cloud storage in Rust using Tokio and Axum to store and stream 80TB+ of build artifacts and OS images for 30,000+ monthly users. Saved over 25x in storage costs compared to AWS S3 and Google Cloud.

Secured multi-tenant servers with an SSH bastion gateway using one-time passwords and SSH key authentication. Plumbed process limits to automatically prevent fork bombs from crashing shared host machines, and inspected outbound traffic to stop network abuse like unauthorized torrents.

Engineered automated virtual machine management using KVM and libvirt. Enabled one-click cloud-init OS installations (Ubuntu, NixOS, OPNsense), dynamic CPU hotplugging, and browser-based VNC access.

Built conversational AI agents that operate directly inside WhatsApp, providing daily users with an instant, zero-friction interface for everyday tasks without having to download separate apps.

Integrated automated tool-calling workflows for everyday utility actions, including live travel ticket checking and phone number identity verification, with sub-50ms query responses.

Architected and deployed self-hosted generative image and video pipelines that let WhatsApp users prompt, generate, and receive media directly inside their chat threads with low turnaround times.

Engineered low-latency microservices in Python, TypeScript, and Rust with persistent state machines, automatically recovering from flaky partner APIs without dropping user session context.

Bridged real-time messaging between WhatsApp and Telegram via custom Matrix Protocol bridges, maintaining reliable zero-message-drop delivery during traffic spikes.

Integrated speech-to-text models from Project Bhashini (co-developed with IIIT Hyderabad) into the clinical patient intake platform, fine-tuning them for regional Telugu accents, rural colloquialisms, and medical vocabulary to reduce transcription error rates by 40%.

Engineered an automated voice ingestion system with voice activity detection and background noise filtering so nurses and doctors could use the system in noisy clinic reception areas.

Built real-time evaluation pipelines that extracted symptoms, duration, and patient severity to generate standardized EHR clinical notes ready for doctor review before appointments.

Developed the backend web application in Python and Django with strict multi-tiered role-based permissions (Employee, Officer, and Chief Officer) to ensure defense communications remained securely compartmentalized.

Built a secure integration with Google Workspace OAuth2 APIs that dynamically created encrypted video meeting links on demand when inter-laboratory collaboration requests were approved.

Containerized the platform with lightweight Alpine Docker images and built automated GitHub Actions CI/CD workflows to build, test, and deploy releases directly onto hosted Oracle Cloud infrastructure.

Wrote custom automation scripts using Expect and custom Django management commands so new server containers could initialize and configure secure admin accounts automatically without manual intervention.