Hire Shreyas M. - AI Engineer

· Architected a fully agentic fraud intelligence platform (LangGraph + MCP) autonomous agents plan, retrieve, invoke live tools, and resolve cases end-to-end; 47% reduction in investigation turnaround and $2.1M+ estimated annual savings across enterprise payment systems. · Designed and owned the LLM evaluation framework (adversarial red-teaming, output scoring, hallucination detection, bias auditing) serving as the production deployment gate for 12+ live LLM workloads, cutting post-release model incidents by 60%. · Built multi-agent orchestration with CrewAI and AutoGen specialist agents run in parallel across anomaly detection, AML compliance, and fraud correlation; 31% reduction in analyst workload and 3× throughput on peak transaction volumes. · Deployed RAG-based AI copilots (LlamaIndex + Pinecone + Elasticsearch hybrid retrieval) average analyst query resolved in under 8 seconds vs. 25+ minutes previously; 94% analyst satisfaction score post-launch. · Implemented LLM fine-tuning pipelines(LoRA, QLoRA, RLHF) domain-adapted models outperform base GPT-4o by 18% F1 on AML classification; 4 models shipped to production in 12 months. · Built AI observability and cost optimisation (W&B + MLflow) reduced monthly LLM inference spend by 34% while maintaining quality SLAs across 5+ production services. · Delivered cloud-native AI infrastructure on AWS (SageMaker, Bedrock, Lambda) with Terraform IaC and GitHub Actions CI/CD zero-downtime deployments across 99.9% system uptime. · Engineered Responsible AI guardrails (PII redaction, output filtering, PCI-DSS audit trails) passed all 4 enterprise AI governance audits; 0 compliance violations since platform launch.

* · Built multilingual RAG pipeline (Hugging Face * FAISS, 10 * languages) serving millions of daily queries 28% search relevance uplift in tier-2/tier-3 cities; sub-100ms P95 latency in production on GCP. · Deployed real-time personalisation engine (Kafka * PySpark * LLM re-ranking) processing 50M * signals/day drove 15% cross-sell conversion increase and contributed to $8M * incremental revenue during peak BBD sale events. · Developed agentic shopping assistants (OpenAI * LangChain tool-use) autonomously handling product discovery, FAQ resolution, and order tracking deflected 40% of inbound support volume to self-service. · Optimised dynamic pricing models(XGBoost * reinforcement learning) 12% GMV growthand 9% improvement in seller margin efficiency across 3 major product categories. · Implementedvisualproductsearch(PyTorchdeeplearning) 91%top-5accuracy; enabledimage-baseddiscovery for 10M * catalog items from smartphone uploads. · Engineered fraud detection models (LightGBM * Scikit-learn * XAI) 22% reduction in false positives, $1.4M in prevented fraud losses across buyer and seller networks in first year. · Designed scalable REST and gRPC microservices (FastAPI) serving all GenAI and ML predictions 99.95% SLA uptime, auto-scaling on GCP Kubernetes Engine handling 10K * RPS at peak.