Saad – Python, React, Microsoft Azure

A productivity platform powering real-time, multilingual communication and task automation through low-code, multi-agent workflows and conversational assistants. Having a drag-and-drop workflow system enabling cross-sector automation (government, banking, HR, customer support), cutting time to completion of business procedures by 50%. Comprising of agentic RAG, MCP server integration, 20+ tools and data source connectors, 10+ LLM provider support and deployed with Arize Phoenix observability on Kubernetes.

• Owned and architected the AI backend platform, a multi-tenant productivity and automation SaaS supporting real-time, multilingual AI assistants and agentic workflows
• Designed scalable, event-driven architectures enabling low-code multi-agent workflows used across government, banking, HR, and customer support domains
• Built customisable drag-and-drop task orchestration for AI workflows, reducing time-to-market by 50% for new enterprise use cases
• Implemented event-driven LLM orchestration using LlamaIndex for agentic RAG and tool execution, improving development velocity by 30% through modular design
• Developed 10+ production-grade AI function tools (file operations, REST APIs, web search, document generation, code execution), enabling agents to perform real-world actions
• Integrated MCP server & A2A to support extensible agent-tool and agent-agent interactions across diverse customer workflows
• Enabled multi-LLM support (Azure AI, OpenAI, self-hosted Ollama), allowing flexible deployment strategies across cost, privacy, and performance constraints
• Optimised RAG pipelines using MongoDB Atlas vector search with hybrid and semantic retrieval, improving search accuracy by 25%
• Delivered real-time Text-to-SQL and Workbook AI engines, integrating data from REST APIs, SharePoint, S3, and Azure Blob Storage, driving a 60% increase in user engagement
• Maintained high-throughput AI production systems handling 10K+ queries/month, achieving 75% error-free operation under peak load. Deployed and operated a self-hosted Arize Phoenix observability platform on Kubernetes, accelerating AI issue root-cause analysis by 80% and enabling continuous performance optimisation

A CMS Compliance Intelligence Engine that helps skilled nursing facility operators respond to survey deficiencies, train staff, and build formal compliance documents. The platform uses AI agents powered by large language models combined with a retrieval-augmented generation (RAG) system that draws from two knowledge bases: real-world CMS-2567 survey citations and official CMS regulatory documents

• Architected and built a full-stack AI compliance platform from scratch using FastAPI, React 19, TypeScript, and Google Vertex AI, serving skilled nursing facility operators with real-time regulatory guidance
• Designed a RAG pipeline processing 101,600+ CMS survey documents and 44 regulatory PDFs through a two-stage retrieval system (vector similarity fetch of 100 candidates → BGE reranker → top 16 results), achieving high-precision citation retrieval across 3,072-dimension embeddings
• Built 4 specialized AI agent personas using LangGraph and Gemini, each with distinct compliance workflows — from triage and incident analysis to staff training generation and formal Plan of Correction drafting with exportable PDF/DOCX output
• Engineered a multi-provider authentication system using Firebase Authentication (Google, Microsoft, Email/Password) with zero-downtime migration from Google OAuth — preserved all existing user IDs, subscription data, and GCS file paths by importing users with original UIDs into Firebase
• Implemented end-to-end Stripe subscription billing with 30-day free trial logic, one-trial-per-user enforcement, webhook-driven lifecycle management (6 event types), automatic downgrade on payment failure, and customer self-service portal
• Built an automated PII redaction pipeline using Google Cloud DLP that processes PDFs (text and scanned), DOCX, and images — detecting and redacting SSNs, phone numbers, names, addresses, and DOBs before storage in user-scoped GCS vaults
• Delivered real-time streaming responses via Server-Sent Events (SSE) with live markdown rendering, tool-use status indicators, and artifact visualization (citations with relevance scores, follow-up suggestions)
• Deployed on GCP infrastructure: Cloud Run (containerized backend), Firebase Hosting (SPA), Google Cloud Storage (private vault), Vertex AI (embeddings + generation), Cloud DLP (PII protection), with Neon PostgreSQL and Pinecone vector DB
• Ingested 101,681 documents into Pinecone using a custom pipeline with chunking (8K chars with 400-char overlap), optional LLM-powered chunk enrichment, batch embedding with rate-limit handling, and idempotent processing via file tracking

A region-sensitive, data-augmented AI companion designed to support autism caregivers in the UK and India, integrating live NHS.UK and medical publication search into an adaptive chat interface

• Built using Python and Streamlit, the system leverages LLM-based reasoning for localized caregiver guidance
• Evaluated through expert reviews of real-world caregiver queries, confirming that simplicity, localization, and trustworthiness are key for effective digital health support
• Set up the product in a way that bridges evidence-based AI with healthcare accessibility
• This product won me the best postgraduate student project of the year

An LLM-powered platform for building customizable AI agents and agentic workflows that connect to MS Teams, SharePoint, OneLake, YouTube, and other private data sources.

• Architected and led a team of engineers to create Data Science Dojo's first SaaS product: An LLM Generative AI application for creating customizable AI agents and agentic workflows to talk to your private data from MS Teams, data from SharePoint, One Lake directories, YouTube videos, Websites and many more
• Set up a db schema using PostgreSQL. Deployed Azure resources with load balancing, monitoring and alert rules. Setup Logic Apps and Python/PS Function Apps for automation and serverless computation. Integrated Front Door and WAF for advanced security. Designed distributed backend systems using Service Bus and Python workers (Celery/Celery-beat)
• Streamlined the deployment and semantic git versioning process with CI/CDs on Azure DevOps for around 10 repositories, thus saving 80% of time spent on deployments and keeping track of changelogs
• Delivered transactable 1-Click Azure Marketplace offer with metered billing and created plans that eventually landed us 2 more customers. This managed application is deployed using an ARM solution template with infra-on-customer tenant, thus ensuring 100% data governance and GDPR compliance
• Migrated infrastructure across Azure subscriptions with zero downtime, saving $20K in cloud costs
• Implemented Django as an OAuth provider with OAuth2.0 protocols by setting up authorization code and client credential flows to authenticate and authorize from different sub-applications
• Integrated automated distribution list creation for agents, leveraging Entra ID Enterprise app to integrate with Microsoft Admin 365, showcasing adeptness in system integration and automation, thus resulting in 90% decrease in manual work
• Implemented Agile, SDLC, and best engineering practices. Wrote software specification document, designed the application infrastructure and prioritized features using MoSCoW analysis

A tailored grants management ecosystem, centered on two interconnected web applications that became the backbone for an international organization’s team. It digitized entire workflows: authorities could submit, review, and approve proposals online, ditching manual paperwork for good. Real-time dashboards tracked every detail such as lifecycles, budgets, beneficiary stats, and financials while automated notifications kept everyone in the loop via email. Organizations could submit proposals, upload reports, and share documents seamlessly. They got instant updates on status, funding, and requirements, building trust through transparency.

• Worked on and led a team of developers for a Grants Management System (GMS) by an international governmental institution. By creating an end-to-end flow of organizational operations along with automation and Power BI visualizations, the GMS was able to save more than 10 hours of manual work for grantees and beneficiaries
• Engaged in deliverable discussions, feedback, and requirement collection with the clients and the beneficiaries. The positive communication, timely delivery, and powerful outcomes resulted in commencing a new multi-regional GMS
• Acted as a consultant for a new phase of GMS and contributed to analytics by creating Power BI dashboards. Also helped in setting up a multi-tenant application and database architecture in ASP .NET Core
• Setup automation in GMS project to email Power BI analytics PDF reports on a biweekly recurrence to specific distribution groups by utilizing Microsoft Power Automate, reducing the time of manual work by 90%