Software Engineer at Arbisoft Lahore, PK

AI, built for
the real
world.

I’m Muhammad Tayyab Tahir Qureshi.

I turn AI capabilities into working products. Agent systems, retrieval pipelines, and the backend engineering that brings it all together.

Explore my work Download CV

Building at Arbisoft. Studying AI at LUMS.

Systems in motionINTERACTIVE STUDY / 2026

An interwoven lavender sculpture representing coordinated agent systems

01 / AGENT SYSTEMS

Many agents. One coordinated system.

Connected loops represent tool use, orchestration, and human oversight.

Explore Nightshift Agents ↗

Visual metaphor · Move your pointer to explore

Building atArbisoftSoftware Engineer · AI Focus

Studying atLUMSMS Artificial Intelligence

Working acrossModels → SystemsAgentic AI · Backend · Cloud

01 / Selected work

Built at Arbisoft & independently

Complex problems.
Considered solutions.

A closer look at what I build, how it works, and the decisions behind it.

NIGHTSHIFT01 / AGENT ORCHESTRATION

Autonomous work.
A human in the loop.

Architecture illustration

01 ARBISOFT

Autonomous workflows / Agent infrastructure

Nightshift Agents

Autonomous work. Human oversight.

Agents that ingest documents and email, extract structured information, propose work, and report to an operator in Slack. Sensitive actions require human approval, with an audit trail across the workflow.

Amazon Bedrock
MCP
n8n
AWS ECS
Terraform

Read engineering notes

The challenge

Coordinate unattended agent workflows while keeping sensitive actions reviewable and operational history traceable.

My contribution

Built scheduled n8n heartbeat and ingestion workflows for Google Drive and Gmail.
Designed a custom MCP gateway with bearer authentication, field-level encryption, and append-only WORM audit logging.
Implemented the Bedrock reasoning loop, adapting MCP tools to the model’s tool-use interface.
Containerized services and provisioned ECS, an internal load balancer, a private VPC, and Secrets Manager with Terraform.

A deliberate boundary

The reasoning loop uses a one-retry-then-escalate policy. Agents propose changes; sensitive actions are gated behind human approval.

Professional work at Arbisoft · Description based on my August 2026 CV.

HELPDESK / AI02 / INTELLIGENT ROUTING

One request.
The right way forward.

Architecture illustration

02 ARBISOFT

Multi-agent orchestration / Retrieval

IT Helpdesk Agent

The right route for every request.

A LangGraph system that classifies requests, routes them through retrieval, integration, or escalation lanes, and restores conversations from PostgreSQL checkpoints.

LangGraph
ChromaDB
FastAPI
PostgreSQL

Read engineering notes

The challenge

Combine knowledge retrieval and tool execution with recoverable conversations and responsive document ingestion.

My contribution

Architected a classifier-driven orchestrator with dedicated subgraphs and namespaced state.
Designed a Chroma-backed retrieval pipeline feeding an LLM tools loop, with PostgreSQL checkpoints.
Implemented single and batch uploads, document lifecycle management, and presigned previews.
Moved processing into Taskiq workers and built streaming chat with Next.js and FastAPI in an Nx monorepo.

Engineering decision

Document processing runs outside the request path. Conversation checkpoints support recovery across sessions.

BENCHPREP03 / ASYNC ENGINEERING

Keep the conversation going.
Let the workers work.

Workflow illustration

03 ARBISOFT

Backend engineering / Real-time sessions

BenchPrep

An interview platform, built for async work.

FastAPI services for interview management, real-time sessions, background AI evaluation, and candidate matching, with secure access to resumes and recordings.

FastAPI
Taskiq / Redis
LiveKit
SQLAlchemy
S3

Read engineering notes

My contribution

Developed async APIs with Pydantic validation, PostgreSQL persistence, and Alembic migrations.
Integrated an OpenAI-powered evaluator running off-request in Taskiq workers.
Implemented LiveKit interview sessions and S3 presigned upload and download URLs.
Built background candidate scoring and matching, plus separate reminder and overdue-detection workers.

Engineering decision

Interview evaluation and scheduled work run separately from interactive API requests, keeping long-running processing out of the request path.

Further explorations

More to the story.

04 INDEPENDENT

Semantic search / Django

Wagtail AI Chatbot

A retrieval and chat plugin for Wagtail content, with persistent indexing, metadata filtering, and a modular LangChain pipeline using FAISS and ChromaDB.

Read implementation notes

Built semantic search and LLM-powered chat over site content. Separated retrieval from generation to support persistent indexes, metadata filters, and scalable vector retrieval within a Django/Wagtail integration.

RAG / LANGCHAIN

05 INDEPENDENT

Computer vision / Perception & control

From pixels to heading

A monocular-video pipeline combining PIDNet segmentation, dense optical flow, and a PD yaw controller to produce heading and speed commands. Includes annotated video and per-frame logs.

Explore the repository

PYTORCH / OPENCV

06 ARBISOFT

Cloud deployment / Applied AI

Sama X · Voice of Customer

A call-review application for analyzing customer call summaries. Long-running inference runs in AWS Lambda, with a FastAPI and Python stack deployed across EC2, Lambda, and S3.

Read implementation notes

Delivered the call-review application and deployed and maintained its AWS stack. Offloaded long-running AI inference to Lambda for event-driven execution.

FASTAPI / AWS

02 / Background

Practice, with a foundation in theory

Curious by nature.
Engineer by practice.

My work sits between machine learning and software engineering: understanding a model, connecting it to useful tools, and building the infrastructure around it.

At Arbisoft, I contribute to AI products and platform engineering. Alongside that work, I’m pursuing an MS in Artificial Intelligence at LUMS.

Read my full CV

JUN 2024 — PRESENT

Software Engineer · AI Focus

Arbisoft

Agent systems, production backends, and AWS deployments. Also contributed to Open edX modernization, including Python 3.12 enablement, deprecation fixes, and platform modularization.

2025 — PRESENT

MS Artificial Intelligence

Lahore University of Management Sciences

Weekend program CGPA 3.88

2020 — 2024

BS Computer Engineering

University of Engineering & Technology, Lahore

CGPA 3.827

03 / Technical repertoire

Tools in service of the system

The tools.
The thinking.

From a retrieval pipeline to the infrastructure it runs on.

01Agents & retrieval

LangGraph · LangChain · MCP · n8n · LLM orchestration · RAG · ChromaDB · FAISS · Amazon Bedrock

02Backend & data

Python · FastAPI · Django · Django REST Framework · PostgreSQL · SQLAlchemy · Alembic · Taskiq · Redis

03Cloud & infrastructure

AWS ECS · EC2 · Lambda · S3 · ECR · VPC · Docker · Docker Compose · Terraform · Git

04Machine learning

PyTorch · scikit-learn · Hugging Face · OpenCV · NumPy · Pandas · PySpark · Transformers · CNNs · NLP · Computer vision

Further explorations

Document QA · CNN image classification · Canny edge detection from scratch · RISC-V processor pipeline

GitHub ↗Spam classifier ↗Movie genre classification ↗

Continued learning

Generative AI — Analytix Camp
Supervised Machine Learning & Advanced Learning Algorithms — DeepLearning.AI
Intro to Machine Learning & Data Visualization — Kaggle

04 / Get in touch

Lahore, Pakistan · UTC +05:00

Let’s build
what’s next.

Have a role, a project, or a challenging AI problem in mind? I’d like to hear about it.

m.tayyabtq@gmail.com

LinkedIn ↗GitHub ↗Kaggle ↗

AI, built forthe realworld.✳

Many agents. One coordinated system.

Complex problems.Considered solutions.

Nightshift Agents

The challenge

My contribution

A deliberate boundary

IT Helpdesk Agent

The challenge

My contribution

Engineering decision

BenchPrep

My contribution

Engineering decision

More to the story.

Wagtail AI Chatbot

From pixels to heading

Sama X · Voice of Customer

Curious by nature.Engineer by practice.

Software Engineer · AI Focus

MS Artificial Intelligence

BS Computer Engineering

The tools.The thinking.

Let’s buildwhat’s next.

AI, built for
the real
world.

Complex problems.
Considered solutions.

Curious by nature.
Engineer by practice.

The tools.
The thinking.

Let’s build
what’s next.