Nitin Prajwal 

• Contributed to the architecture of a multi-tenant Cloud Security Posture Management (CSPM) platform on OCI, tasked with monitoring millions of cloud assets for compliance drift.
• Engineered a core module for the platform's policy-as-code (PaC) engine in Go, enabling security teams to programmatically define and enforce custom compliance rules.
• Optimized a distributed data ingestion pipeline using Kafka and gRPC, successfully reducing the 99th percentile processing latency for configuration data streams to under 250ms.

• Developed and led a 3-part technical workshop series on "Building RAG Systems," covering vectorization, indexing strategies, and LLM orchestration, attended by over 150 students.
• Directed the chapter's Google Solution Challenge project, creating an AI tool that fine-tuned a BERT model for extractive summarization of dense municipal policy documents, increasing accessibility for local journalists and advocates (SDG 16).
• Collaborated with industry professionals on projects utilizing natural language processing and search technologies, building expertise in core Google technical areas.

• Developed a surrogate AI model using a Message Passing Neural Network (MPNN) to approximate computationally expensive Density Functional Theory (DFT) simulations.
• This model reduced the simulation runtime for new carbon-capture material candidates from weeks to hours, achieving a 92% predictive accuracy (R²) and significantly accelerating the R&D feedback loop.
• Implemented distributed computing solutions to handle large datasets, demonstrating ability to research and develop systems for information retrieval at scale.

• Designed and implemented an LSTM-based autoencoder to detect anomalies in satellite telemetry data, providing a 72-hour advance warning for component degradation with a 99.5% F1-score.
• Built the supporting serverless ETL pipeline on AWS (Lambda, Glue) that processed, cleaned, and vectorized terabytes of time-series data, improving the reliability of a critical satellite subsystem.

Implemented full landing Page for the client with RazorPay integration and N8N backend to servers

• Implemented the FreeRTOS kernel on a custom ESP32 board, using task prioritization and memory protection to resolve 99% of race conditions and device hangs from the previous bare-metal firmware.
• Enhanced device security by integrating an ATECC608A crypto-element for secure key storage, preventing firmware cloning across a fleet of 1,000+ deployed educational kits.