Experience

My contribution

I built a LangGraph sales copilot integrating Salesforce, Gmail and other business tools, with lead qualification and Slack drafts for human review. I developed the Salesforce MCP server in Python with FastMCP so the agent could retrieve customer details, search opportunities, create and update leads or opportunities, summarise interactions and create support tickets.

Actions followed the requesting user's existing roles and permissions. Alongside the integration, I built and customised the agent's execution loops, tool routing, workflow state, context and memory, failure recovery, evaluation and human-approval behaviour. I deployed the product on AWS with infrastructure managed through Terraform.

Outcome

Sales representatives, account managers and leadership or operations users used the system in production. It reduced manual CRM entry and updating and let users complete several CRM actions in one conversation.

My contribution

I collaborated on the design and development of a portal bringing together engagement updates, action ownership and answers grounded in previous meetings. It gave users a place to review progress and retrieve meeting context with references.

The portal also supported demonstrations to prospective clients.

Outcome

Demonstrations contributed to signed new client contracts.

My contribution

I built LangGraph/Gemini agents for web chat and WhatsApp covering availability, bookings, scheduling changes, cancellations, refunds and payments.

I also developed a telephone agent that handled complete booking calls for courts and training sessions across football, basketball, padel and tennis facilities in Saudi Arabia.

For managers, I built natural-language analytics that translated questions into interactive React charts covering revenue, bookings, payment methods, staffing and trainer performance.

Reliability and accessibility work included Arabic support, image and voice-note processing, persistent memory, guardrails, human handover and automated LangSmith evaluations.

Outcome

The web/WhatsApp agents, telephone booking agent and conversational analytics were used by real users in production. User feedback provided qualitative evidence of value.

My contribution

I led the project from initial concept through production launch, coordinating engineering and business stakeholders and translating customer needs into delivery priorities.

The implementation combined multi-LLM orchestration, specialised tools and retrieval-augmented generation. I worked on vector search, knowledge graphs and semantic routing, integrated speech transcription and translation, and owned backend deployment and maintenance on AWS ECS.

I led CI/CD delivery, evolved conversation and document storage using PostgreSQL with vector search, and contributed to product discussions, customer engagement, pitches and fundraising preparation.

Outcome

The project progressed from concept to a production product supporting company-knowledge answers and automated customer conversations.

My contribution

I assessed those needs and worked backwards to define how an AI-enabled employee portal could address them. With the project team, I translated the findings into use cases, user journeys, access and governance requirements, and a portal blueprint.

The architecture recommendations covered Microsoft/Azure services, single sign-on, AI-assisted document search, personalised content and enterprise integrations. The implementation roadmap connected proposed capabilities to organisational priorities.

Outcome

The blueprint and roadmap informed prioritisation of the next phase.

On-premise Mistral/RAG pilot

I translated confidentiality and internal hosting requirements into a self-hosted RAG solution using Mistral models on the client's infrastructure. Document retrieval and LLM processing remained within that environment. The pilot was used by selected client users.

FridayGPT cloud-product contributions

Separately, I contributed to FridayGPT's enterprise knowledge assistant in a product used by customers. The work combined keyword, vector and semantic retrieval through Azure AI Search with source-referenced answers.

I improved document preparation through email preprocessing and PDF-to-Markdown extraction with Azure Document Intelligence and contributed Python/FastAPI services for chat, embeddings and document question answering.

Outcome

The on-premise work reached a selected-user pilot. The cloud contributions were deployed in the customer-used product.