Ravi Narayan Mahawar

• Shipped Broker After-Hour Referral, letting brokers get Managing Partner sign-off outside business hours to fast-track binds instead of losing the sale to the appeal process. Drove 1,000+ referrals in two weeks and CAD 6M+ in business retained in the first two months.
• Built a runtime policy-bind evaluation engine: an LLM agentic workflow in LangGraph, grounded in the underwriting guidelines through RAG, that flags plausible issues at bind time so brokers correct errors before exposure reaches audit.
• Engineered a policy auditing platform running 111+ validations across underwriting, soft-fraud and premium-leakage to surface pricing exposure on bound policies.
• Launched an AI-driven quoting channel through a ChatGPT-native interface, letting customers get preliminary quotes inside an AI assistant.
• Led a caching layer precomputing the Sales, Service and Book of Business dashboards into mathematically derived buckets, cutting load times from seconds to under one second with separate admin and broker views.
• Built an MCP (Model Context Protocol) server and Claude agent orchestration tooling giving internal agents safe access to company data and automating developer workflows.

• Developed an open-source intelligence (OSINT) monitoring system to aid in identifying potential arms embargo violations and support global disarmament efforts.
• Implemented a global web scraper using Crawl4AI that performs shallow digs for relevant data, escalates to deeper exploration on human approval, and uses LLMs to summarize findings and surface geolocated resources worldwide.
• Made new resources discoverable for the team while keeping a human in the loop for source reliability, eliminating manual repetitive work across data ingestion, context mapping and association.

• Improved customer onboarding efficiency by 60% with a dynamic salary breakdown service for 40+ countries and currencies.
• Reduced processing latency by 40% by integrating Amazon MSK and establishing 10+ asynchronous channels between microservices.
• Mentored junior engineers and led the extraction of microservices from the existing monolith, building the new services from scratch.

• Drove a 20% increase in bookings, adding 500 room nights daily, by leading the Day Wise Promotion project.
• Spearheaded adoption of closed user group discounts, achieving CUG application in 50% of all bookings.
• Migrated three TIER-1 services from ECS to EKS.

• Increased API performance by 40% across 10+ user-facing endpoints and achieved 99.99% uptime for payment services by integrating Razorpay.
• Deployed an auto-scaling Celery system handling 150+ parallel tasks per day.

• Built a caching mechanism with periodic refresh and schema validation, eliminating a database hit on every request, plus an S3-backed fallback data source for availability.

• Founded an independent AI/ML research consultancy taking on applied research engagements for clients under NDA.
• Researched task hardening for agents, training them on difficult problems with reinforcement learning and other applied ML methods.
• Designed multi-agent Claude pipelines that simulate and regress on hardened tasks, measuring them against target outcomes and fail rates.
• That pipeline became the client's foundational multi-agent swarm workflow, taken up by other contributors and adopted across all of their projects.

• Co-founded a multidisciplinary non-profit research lab with Dr. Shohini Ghose and Dr. Toorjo Ghose.
• Research projects on combinatorial mathematics, mental health datasets, and photon light activity correlation.
• Research directions intersect mathematics, quantum computing, AI/ML and high-performance computing.
• Collaborators include Dr. Ilias Kotsireas and Dr. Nirosha Murugan.
• Models and datasets are held private under IP protection pending publication of the associated papers.

• Developed optimized solutions for D-optimal, Legendre Pair and Hadamard Matrix problems using serial and MPI implementations, testing performance prior to quantum resource implementation.
• Designed reinforcement learning algorithms that solved combinatorial problems at two-digit problem sizes, then reached three-digit sizes by transferring the learned policies rather than searching from scratch.
• Engineered three quantum algorithms with D-Wave, focusing on performance optimization and resource efficiency.
• Built an explainable MARL framework on top of VIPER, evaluating performance across 3 games and 40+ parameter combinations on AWS.

Waterloo, Canada

Bhubaneswar, India