Raju R.

About Raju R.

Lead AI Engineer with expertise in cloud native infrastructure and AI driven operational intelligence, integrating LLM based workflows and AI automation platforms to improve reliability, performance and engineering productivity.Achievements as Individual Contributor:Developed an LLM powered semantic data layer to auto classify schemas, generate business metadata and map raw data to KPIs and governance policies.Optimized feature engineering and ML workflows by prioritizing high impact signals in feature stores, doubling model deployment velocity and improving predictive accuracy by 25%.Developed cost aware orchestration logic leveraging cloud billing APIs and workload telemetry to balance performance, reliability and spend in real time.Implemented RAG tenant isolation, ensuring retrieval is scoped to authorized groups and preventing cross tenant leakage via embedding based search.Designed scalable pipeline execution across scheduler worker architecture with serialized dag metadata stored in a meta store and coordinated via an api server gateway. Designed a cloud native autonomous data platform using kubernetes, kafka, spark and delta lake to dynamically optimize data ingestion, processing and storage based on AI driven value scoring.Designed and implemented an AI based data value engine that scores datasets and fields in real time using business KPI correlation, predictive signal strength, cost to store and compliance risk.Built a distributed tracing model that represents a complete agent run as a single trace with spans for LLM inference, embeddings and retrieval, tool execution and post run evaluation.Built an oltp pipeline to export traces,metrics, events through the open telemetry collector for centralized processing, enrichment and routing to observability back ends.Optimized inference latency by implementing a tiered caching strategy for the symbolic layer, ensuring that recurring intent patterns are resolved in sub 100 ms without redundant LLM calls.Designed a schema agnostic mapping layer that allows the platform to support breaking upstream API changes without impacting end user functionality through automated semantic rerouting.Implemented asynchronous event driven orchestration by deferring external wait tasks to the triggerer, releasing worker capacity for runnable tasks and improving overall throughput.Executed iceberg table maintenance to control metadata growth and performance: snapshot expiration, datafile rewrite, compaction and manifest management to reduce runtime file open overhead.

Experience

1. Lead Ai Engineer

   Autodesk

   Apr 2025 — Present · Bengaluru, IN

   Built a cloud native AI platform that does not just move data, it decides what data matters based on business impact, cost and compliance risk.Designed an AI driven control plane that scores datasets in real time and routes them across hot, warm and cold pipelines automatically.Implemented LLM powered semantic mapping to link raw data directly to KPIs, ML models and decision workflows.Implemented bi encoder retrieval with cross encoder re ranking to boost relevance while managing compute cost tradeoffs in production search pipelines. Deployed autonomous governance with real time PII detection and adaptive access controls that:Reduced data infrastructure costs by 40%.Improved analytics performance by 35%.Boosted ML model accuracy by 20%.This shifted data from a cost center into a self optimizing business asset.Built semantic search scoring pipelines using embedding similarity and multiple distance metrics enabling graded relevance beyond keyword overlap.Designed knowledge graphs by defining ontology extracting triples from unstructured text, normalizing entities and applying confidence scoring prior to graph database storage.Implemented out of sample embedding for newly arriving nodes using adjacency vectors and graph vertex removal utilities, enabling inductive updates without retraining from scratch.Built vertex ranking systems using seed nodes and spectral embeddings, prioritizing high risk nodes for investigation rather than only binary classification.Designed parsimonious model selection using sequentially nested hypotheses and likelihood based testing to avoid overfitting while escalating model complexity only when justified by evidence.Optimized large scale graph computations via sparse matrix storage and truncated svd, achieving order of magnitude runtime improvements over dense decompositions for spectral pipelines.

Education

• International Institute of Information Technology Bangalore

  M Tech Computer Science Engineering CGPA 9.95 out of 10, Rank 1, GATE 2011 CS AIR 28

• Indian Institute of Science (IISc)

  Post Graduate Level Advanced Certification, AI ML Ops : Grade A+

• Shanmugha Arts, Science, Technology and Research Academy

  B Tech CGPA 8.86 First Class with Distinction, Batch Topper, AIEEE Qualified