- Architected and deployed a supervisor-led multi-agent system to automate complex enterprise HR workflows, utilizing a "Supervisor Agent" to intelligently route tasks to specialized agents (Leave, Policy, Employee Info). - Built an onboarding agent that handles the entire new hire process, from verifying IDs using OCR, securely collecting banking information, to setting up work hardware, streamlining the first day for new employees.

- Built autonomous agents using LangGraph and LangChain capable of multi-step reasoning (ReAct, Chain-of-Thought) to handle leave applications, performance evaluation, policy discovery, and PII updates across Microsoft Teams, Web, and mobile. - Designed a hybrid retrieval layer using semantic and lexical search(Pinecone) and PostgreSQL for structured data, integrated with custom Model Context Protocol (MCP) servers for real-time organizational data access.

- Implemented sophisticated context engineering and memory management within an asynchronous FastAPI backend, ensuring high-throughput performance and coherent multi-turn conversations. - Established an enterprise-grade MLOps stack using LangFuse (for observability) and RAGAS (for measuring metrics such as Faithfulness and context relevance), deploying a self-hosted Dockerized architecture that reduced hallucination rates by 35% and accelerated task completion speed by 60%.

Designed and deployed a domain-optimized LLM pipeline (RAG), reducing lithium brine analysis time.

Built scalable back-end services on AWS (App Runner, S3, CodeBuild, SQS) with Docker, ensuring secure, automated deployments and high availability.

Implemented hybrid data management with PostgreSQL (structured data) and Pinecone (Vector DB) to enable context-aware semantic search.

- Architected and built a scalable FastAPI backend with MongoDB, JWT auth, RBAC, and async logic for high throughput. - Engineered mobile money payment integrations with retry logic, real-time updates, and async background processing (Cloud Tasks, Pub/Sub, Scheduler, Run, Storage bucket, etc).

- Designed indexed MongoDB schemas and optimized queries to support 1,000+ concurrent payment transactions. - Built secure REST APIs for internal services, third-party integrations, and USSD flows for offline users.

- Automated deployments with Docker and GitHub Actions CI/CD on Google Cloud Platform, integrating monitoring, logging, and alerting. - Managed the full backend lifecycle, from authentication to webhook orchestration, ensuring seamless recurring payments with a 98% renewal success rate and <1% churn due to payment failures.

Architected a web widget + WhatsApp assistant, enabling customers to search and shop using natural language.

Built a multimodal LLM pipeline to extract attributes when brand upload outfits and drive semantic search across 2K+ catalog items.

Integrated HeyBeauty's 2D avatar engine, allowing users to preview recommended outfits on themselves with real-time virtual try-ons.

Delivered scalable, reliable AI-powered knowledge workflows by developing Besti, an LLM chatbot platform, overseeing Python back-end and AI architecture for enterprise usage.

Integrated multiple data sources, including documents, PDFs, websites, sitemaps, and cloud storage, to provide context-rich and accurate insights.

Built scalable FastAPI services with real-time streaming via Pusher, containerized with Docker, and deployed to GCP Cloud Run with automated monitoring dashboards in Grafana.