STAFF ENGINEER / PROPTECH + AI
ClearSplit01
From zero to a property platform. Revenue splits, trustworthy ledgers, and AI onboarding with a human in the loop.
$6.6M+ transaction volume4,200+ reservations · first 30 days live
STAFF ENGINEER · CLEARSPLIT · DUBAI
Serious code.
Curious mind.
I’m Mahad. I build platforms from the ground up,
and AI agents that turn complex workflows into useful work.
Product thinking. Deep engineering. A little curiosity.
Give one a job. ↗
THE SYSTEMS PLAYGROUNDFIG. 001
01Product thinkingFROM ZERO TO ONE
CLOUD + AI ORCHESTRATIONSystems that scale02
A SYSTEM, IN THREE DIMENSIONS
+ One product. Many connected layers.+
THERE’S MORE BELOW
Interfaces✳APIs✳Cloud✳AI agents
BUILDING PLATFORMS & AGENTSNext.jsLangGraphTypeScript.awsLangChainPostgreSQL
01 / SELECTED WORK
Built for the
real world.
Behind every interface, there’s a system.
Here are a few I’ve helped bring to life.
COMMERCE / SERVERLESS
Liverecover02
Turning abandoned carts into conversations. Event-driven infrastructure for the moments that matter.
From webhook to conversationServerless commerce · SMS + email
DOCUMENT AI / REAL-TIME SYSTEMS
AslasChat03
Your documents, put to work. A chatbot platform that connects business knowledge to useful, real-time answers.
Knowledge, in conversationDocument training · NestJS + LLM APIs
More problems solved. More things shipped.
02 / THE TOOLKIT
The right tools.
Connected.
From an interface to an agent that can act.
The useful part is how the pieces work together.
AGENTIC AITOOLS, WITH INTENTION
Agents with a job to do: the right tools, a stateful workflow, and human review before important changes. Built into real products.
LangChainLangGraphTool-calling LLMsDAG workflowsHuman-in-the-loopStateful orchestrationKnowledge retrievalStaged bulk commits
THE FOUNDATION BENEATH THE AGENT.INTERACTIVE SCHEMATIC
01InterfaceNext.js
02APINode.js
03ComputeAWS Lambda
04DataPostgreSQL
Small pieces. Bigger possibilities.
THE AGENT LAB
Give an agent a job.
Watch the system work.
AI gets interesting when it can do useful work.
I build the systems that turn a request into a plan, a tool call, and a result you can trust.
MAHAD’S WORKBENCH / 002A DIRECTED ACYCLIC GRAPH
THE JOB
“Onboard these 100 properties.”
Check references. Prepare changes. Ask first.Preparing the workbench
INPUT → PLAN → TOOLS → REVIEW → COMMIT
Drag to rotate · select a station
READY WHEN YOU ARE
A job is waiting.
INSIDE STATION 02
Directed acyclic graphMake a plan
Dependencies define the route. Two independent tools can run together; review must wait for both. This example has no cycles. LangGraph can also support loops and richer stateful flows.
intake → plan → [lookup, validate]Working Complete Needs a person
Illustrative local simulation. No real property data, AI calls, or external writes.
BUILT FROM REAL WORKInspired by the agentic onboarding systems I build at ClearSplit.
100-property onboarding
30 minunder 2 min
03 / A little about me
ENGINEER'S FIELD NOTES
Understand the problem, the data, and the decisions that still need a person.
From a blank page to something people rely on.
I'm Mahad, a Staff Engineer in Dubai. I like understanding the whole problem — then taking responsibility for making it work.
At ClearSplit, I built the platform from zero: the product, the profit-split engine, the integrations, and the infrastructure behind it. Those systems processed $6.6M+ in transaction volume in their first 30 days live.
Now I also build agents that do useful work. LangChain and LangGraph connect the tools and workflow; review gates keep people involved in the decisions. At ClearSplit, that brought onboarding for 100 properties from 30 minutes to under two.
- Years of experience
- 8+
- Volume in first 30 days
- $6.6M+
- Markets at ClearSplit
- 5+
BS Computer Science
COMSATS University, Pakistan · 2018–202204 / Experience, in practice
A few stops
along the way.
From backend services to founding a platform.
More ownership. A wider view of the system.
Built the platform from zero as the founding engineer: profit splitting, revenue sync, transaction processing, and owner statements across 5+ markets.
Shipped Hostaway-integrated OTA/PMS pipelines that processed 4,200+ reservations and $6.6M+ in transaction volume in the first 30 days live.
Built agentic onboarding with LangChain tool calling, human review gates, DAG workflows, and staged bulk commits, reducing 100-property onboarding from 30 minutes to under 2 minutes.
Integrated LangGraph for stateful, multi-step orchestration across onboarding and data reconciliation, alongside trust accounting, audit logs, and VAT/WPS compliance infrastructure.
Owned full-stack infrastructure, observability, and CI/CD with 48–72 hour feature cycles; co-led product positioning and go-to-market work contributing to $180K in early funding.
Node.jsNext.jsTypeScriptAWSPostgreSQLLangChainLangGraphTool callingDAG workflowsHuman-in-the-loop
Owned the customer data platform and event-driven messaging services from architecture through on-call support, with webhook ingestion handling up to 150K events per second at peak.
Scaled daily messaging from 5K to 30K conversations as daily revenue grew from $2K to $7K; built Next.js dashboards for real-time messaging analytics.
Designed Aurora multi-AZ infrastructure with read replicas and resolved PostgreSQL connection exhaustion using RDS Proxy and PgBouncer.
Built retries, dead-letter queue replay, idempotency, and rate limits around Twilio and Infobip integrations; stabilized services as primary on-call engineer during BFCM.
Reduced monthly AWS costs from $24K to $15K through Aurora Serverless migration, right-sizing, replica tuning, and log optimization.
Added LangChain-powered knowledge retrieval for AI-assisted responses and GitHub Actions pipelines with test gates, production approvals, and rollback controls.
Next.jsNode.jsTypeScriptAWS LambdaEventBridgeSQSAuroraRDS ProxyLangChainGitHub Actions
Owned backend architecture for AslaseChat, an AI customer service platform built with NestJS and deployed on AWS ECS with autoscaling and CloudWatch monitoring.
Integrated LLM document-training workflows so clients could train chatbots on their files, documents, and images.
Reduced PostgreSQL query execution from 2 minutes to under 1 second through indexing and query refactoring; built backend systems supporting ISO and R2 process requirements.
Authored RFCs, led structured design reviews and technical interviews, and contributed to engineering hiring decisions.
NestJSVue.jsPostgreSQLAWS ECSCloudWatchLLMs
Architected the backend gateway for SemiConnect, bringing multiple digital out-of-home providers into one control layer.
Designed a hybrid data architecture with PostgreSQL for transactions and MongoDB for IoT event ingestion.
Improved API performance by 25% with Redis caching, plus rate limiting and throttling controls.
NestJSPostgreSQLMongoDBRedisAWSIoT
Developed Node.js and NestJS backend architectures for multiple client platforms across AWS environments.
Implemented role-based authentication and authorization, and managed production deployments and environment configurations.
Node.jsNestJSAWSRBACProduction deployments
Delivered end-to-end platforms including Heartsybox and a coach-athlete application, with NestJS, MySQL, Firebase, Flutter, and integrated payments.
Built GraphQL and gRPC microservice communication and secure admin systems with role-based access for clients including Dr Joe Dispenza.
NestJSMySQLFirebaseFlutterGraphQLgRPC
Contributed to distributed microservices using RabbitMQ and gRPC for the Germany-based platform.
MicroservicesRabbitMQgRPC
A TINY DEBUGGING PLAYGROUND
One last thing.
Try breaking it.
Everything is connected. For now.
05 / LET’S TALK
Have a good
problem?
Let’s solve it.
An ambitious idea. A system that needs to scale.
Something that could simply work better.
A CONVERSATION STARTS HERE
GOOD SOFTWARE STARTS
WITH A GOOD CONVERSATION.
WITH A GOOD CONVERSATION.