AY

Loading portfolio

AYAsadullah Yousaf
LHR · PK
AI Engineer | Full-Stack Builder2+ years

I build healthcare workflows, RAG pipelines, and agentic systems — from idea to production — with clinical safety, accurate retrieval, and architecture that holds up as usage grows.

Explore selected work
Start a conversation
AI Eng.
Current role at HospitALL, Lahore
4+
Projects delivered
HIPAA
Aligned healthcare AI systems

Lahore, Pakistan

Healthcare AI · RAG · Agentic systems

01Selected Work

Systems that carry real product weight.

Healthcare AI, hospital operations, generative CAD, and multilingual RAG — shipped with production constraints in mind.

Featured · 01Healthcare AI · Agentic triage

MedIQ

The problem
Clinics need faster symptom intake and safer patient routing, but manual triage overloads staff and increases the risk of missed red-flag cases without structured oversight.
My role
AI engineer building agentic orchestration, clinical oversight points, and secure client APIs.
Key challenge
Keeping triage and routing reliable while preserving physician-in-the-loop validation at high-risk decision points.
Result
Reduced manual triage load, more consistent specialist routing, and a safer deployment pattern with doctor feedback loops and monitoring.
MastraTypeScriptREST APIRBACDockerAWS
Open case study
01 / System map

intake → triage → routing → booking

RBAC · encrypted APIs · physician oversight

More work
02Process

From a production problem to a launched system.

How I work — written so a non-technical client can follow it, not just another engineer.

  1. 01Step

    Understand the problem

    Start with the real workflow pressure — what's slow, unsafe, or expensive. Map users, constraints, and success criteria before touching models.

  2. 02Step

    Define the product contract

    Agree on inputs, outputs, roles, permissions, and the states that matter. Make agent and API boundaries explicit so behavior is predictable.

  3. 03Step

    Design data & retrieval

    Sources, embeddings, schemas, and evaluation sets — designed around the paths the product actually exercises, not toy demos.

  1. 04Step

    Build the system

    Implement agents, APIs, and UI with clean boundaries between request handling, model behavior, and third-party services.

  2. 05Step

    Harden & oversee

    Exercise failure paths, RBAC edge cases, and human-in-the-loop checkpoints. Make sure high-risk decisions never fail silently.

  3. 06Step

    Launch & improve

    Ship with monitoring, gather doctor or user feedback, and tighten retrieval, prompts, and routing as real usage grows.

03About

I work on the parts that have to stay correct.

An AI engineer who likes systems you can reason about — not clever demos you're afraid to ship.

I work on full-stack AI products that have to stay correct under real constraints: patient routing, retrieval quality, secure APIs, and physician-in-the-loop oversight.

My work is pragmatic. I care about failure paths, explicit permissions, retrieval that cites the right context, and code the next engineer can reason about without fear.

Most of my work lives where models meet product constraints — triage accuracy, retrieval grounding, encrypted APIs, and oversight loops — and where the cost of getting it wrong shows up in production. That's the part I find worth doing well.

04Experience

Building AI systems under real delivery pressure.

From generative AI research internships to owning healthcare triage workflows in production.

  1. 2026 — Present

    Current

    Lahore, Pakistan · Healthcare AI

    Artificial Intelligence Engineer

    HospitALL

    AI-powered triage, doctor recommendation, and appointment workflows for clinical web and mobile products.

    • Developed AI-powered triage workflows for patient assessment.
    • Built doctor recommendation and critical red-flag labeling logic.
    • Integrated appointment scheduling into web and mobile applications.
    • Improved workflow behavior through doctor feedback loops.
    MastraTypeScriptAgentsRESTRBAC

  2. Jul 2024 — Aug 2024

    Islamabad, Pakistan · Generative AI

    Generative AI Intern

    Center of Excellence-AI

    AI-assisted writing for early learners, including dataset creation and model fine-tuning.

    • Built an AI-assisted writing app for early learners.
    • Created a 60,000-image handwriting dataset.
    • Fine-tuned MobileNet and EfficientNet for handwriting recognition.
    PythonMobileNetEfficientNetComputer Vision
05Capabilities

What I can actually build and own.

Not a wall of logos. The value I create, the relevant tools, and where I've shipped each one.

Agentic Workflows

Multi-step agents for triage, routing, support automation, and internal operations with clear handoffs.

What I build

  • Multi-step agent orchestration
  • Workflow automation and handoffs
  • Physician / human-in-the-loop checkpoints

Technology

MastraLangGraphTypeScriptPython
Example
MedIQ agentic medical triage
Ownership
Owns agent behavior from intake to routing
06Additional Work

More systems I've built.

Research and adjacent work — smaller in scope, still real systems with real evaluation.

Research · Sequence models

Story Generation: LSTM vs GRU

Comparative pipeline for outline-to-story generation with bidirectional LSTM/GRU baselines and human evaluation.

PythonKerasLSTMGRU
Research · IEEE IBCAST 2025

Floor Plan NLP Research

Final-year research on generating floor plans from natural language, tied to the FloorGen AI product pipeline.

NLPFLUXComputer Vision
07Proof

What I bring, stated honestly.

Domains shipped

Healthcare · RAG · Agentic AI · Generative design

Current focus

AI Engineer · HospitALL, Lahore

Systems depth

Mastra · LangChain · FastAPI · React · Docker · AWS/GCP

Education

BS Artificial Intelligence · Bahria University · CGPA 3.65

Open to focused AI collaborations
08Contact

Let's build something that ships.

If you need healthcare AI, RAG, agentic workflows, or full-stack AI integration — I typically reply within 24 hours.

Email

Location
Lahore, Pakistan
Status
Open to focused AI collaborations

Elsewhere

LinkedIn GitHub