Srihari Janadri

AI and Backend Engineer with 3 years at Mercedes-Benz R&D – shipped production LLM orchestration and MCP-integrated tool pipelines for a voice-driven connected vehicle assistant. Experienced in agentic AI system design, RAG pipelines, and distributed backend architecture across Python, Java, and C++.

Software Engineer at Mercedes-Benz R&D India (2022-10 – Present)

Worked on production-grade cloud-side AI orchestration for a voice-driven in-car assistant and vehicle-side gateway services.

• Collaborated on production-grade cloud-side AI orchestration for a voice-driven in-car assistant at Mercedes-Benz – architecting the full pipeline from LLM prompt construction and tool discovery to invocation and response formatting across vehicle app services including climate, navigation, media playback, ambient lighting, and seat controls.
• Collaborated on Gateway Server design – managing vehicle session lifecycle, tool registration sync, persistent WebSocket-based tool call routing to connected vehicles, and heartbeat-driven disconnect detection to maintain accurate real-time capability state.
• Owned the MCP Client as a live vehicle capability broker – maintaining real-time tool registry synced from Gateway Server events, normalizing vehicle MCP Server schemas, and translating LLM tool calls into vehicle-compatible format before dispatch.
• Designed a multi-gate tool filtering pipeline in the orchestration layer – enforcing availability, session scope, and JSON Schema validation before the LLM sees any tool, reducing hallucination risk by ensuring the model selects only from valid, reachable tools with well-formed arguments.
• Collaborated on failure classification and recovery across cloud-to-vehicle flows, distinguishing vehicle errors, timeouts, and unreachable states, and routing non-critical commands to a TTL-based deferred retry queue on vehicle reconnect.
• Contributed to a vehicle-side C++ gateway serving as the central integration point for OTA update distribution, log ingestion, and cloud event relay across connected vehicle services.
• Implemented C++ file transfer workflows using HTTP/cURL for OTA delivery with gRPC-based IPC for cross-service coordination – introduced job queues to decouple transfer processing from the request path, reducing blocking by ~40%.
• Added retry-with-exponential-backoff for transient network failures, reducing failed transfer attempts reaching upstream services by ~35%.
• Contributed to a Java/Spring Boot ingestion service consuming vehicle signals and publishing to Kafka topics – defined topic partitioning and consumer group strategy enabling downstream teams to consume events independently at scale.

B.E. in Information Science and Engineering – JSS Science and Technology University (2018-08 – 2022-08)