Muhammad Uzair

• Replaced passwords with a device-bound identity: partner services authenticate people, approve transactions and collect signatures through one integration, and the signing key never leaves the person's phone.
• Turned identity verification from minutes of manual document checking per applicant into an automated pipeline: web automation of the national identity portal with English and Urdu OCR.
• Designed and shipped the partner-facing API with per-partner key exchange, and the integration guide that partner engineering teams built against.
• Packaged the platform for on-premise deployment so banks with strict hosting rules could adopt it: the same services as containers with self-hosted databases and queues in place of managed AWS.

My part. Owned the backend and the AWS architecture across five event-driven serverless services, and re-architected the core when the requirements changed.

• Cut the AWS bill by 45%, about $23,000 a year: moved the production database off end-of-life Aurora Serverless v1 to v2 with one to five minutes of downtime and no application change, then consolidated servers. Written up as a three-part series on dev.to.
• Gave supervisors plain-language candidate search: an LLM reads each résumé into structured data in a vector database, so a plain sentence finds candidates that rigid filter forms missed. Adopted across client organisations.
• Made payroll and billing safe to retry: a rerun or a redelivered message never pays a worker twice or charges a client twice.
• Shipped three ways for workers to clock in, including fingerprint hardware shared across tenants and a PIN route for workers without a phone, and removed the per-client tablet purchase that slowed onboarding.

My part. Led backend engineering on AWS: AppSync GraphQL over Lambda services in CDK, a master-pool Aurora PostgreSQL with a database per tenant. Hired one engineer and mentored two.

• Built the wearable data platform on TimescaleDB: up to 15,000 samples per person per day from Apple Watch, Garmin, Oura, Fitbit and WHOOP, aggregated correctly across time zones, with storage cut by up to 90%. It shipped and carried real users' health data.
• Chose build over buy for the realtime data path after evaluating six options: hourly granularity and local raw data the vendor could not offer, and about $20,000 a year in vendor fees avoided.
• Connected medical records through a FHIR aggregator covering Epic, Cerner and Athenahealth networks, so a person's records and device data sit on one timeline.
• Made the AI assistant safe to operate: spend capped per person and per service, every request attributable, and a fault in the limiter can never take the product down; right to erasure across four data stores.

My part. Backend engineer across the Node/Express API and the React Native client; built the time-series platform and the wearable ingestion path.

• Parents speak; the app turns it into plain-language records, next steps and a weekly recap, and their doctor sees a structured view of the child between appointments.
• Grounded clinician chat that cites the child's own records and peer-reviewed literature, with prompt-injection and scope guardrails and rate limiting: an answer a doctor can rely on inside a fifteen-minute appointment.
• One provider layer across Anthropic and OpenAI for streaming, retries, overload deferral and kill switches, so a model swap or a cost limit is a one-line change rather than a rewrite.
• Built for daily releases: in-flight AI work finishes before a deploy, a slow provider cannot slow the rest of the app, and stranded jobs resume on their own.

My part. Built the AI layer end to end and most of the parent app's server: voice log, follow-up tasks, milestones, resource library, summaries, weekly recap, email verification.

• Every finished call becomes a booked job in the customer's CRM with nobody re-keying it: a typed integration that surfaces vendor failures as clear errors, retries with backoff and keeps credentials out of logs.
• Twilio carries the call, Retell runs speech, the product owns the LLM conversation with tool calling; fixed a production defect in the call-transfer path.

My part. Built the CRM integration service.

• Payments end to end on Stripe: wallets, payouts, transactions, commission and accounting, with webhook handling that keeps every balance right.
• Live sessions on Agora with real-time updates over Socket.IO and Redis; background work on BullMQ with a dedicated session worker.
• Shipped with CI, database migrations and a deploy path the client runs without me, plus most of the React admin dashboard.

My part. Built and ran the backend for a freelance client: Express on PostgreSQL with Prisma, Redis, S3.