Muhammad Shahwaiz

Muhammad Shahwaiz AI Systems & Backend Engineering

AI systems. Clear evidence. Human judgment.

I'm a final-year Software Engineering student and co-founder of Menzync. I build Python and TypeScript backends, AI integrations and automated tests, with a particular interest in systems that make their evidence and limitations clear.

View selected work Let's talk
Based in
Sialkot, Pakistan
Time zone
PKT · UTC+5
Currently
Final year, BS SE at UMT
  1. EvidenceNextAct Show what was checked, what contradicts it, and what couldn't be verified.
  2. GroundingEmoSense AI Hold model output to a schema or to retrieved sources, and fall back in the open when that fails.
  3. Human approvalClientDesk Draft, don't send. A person approves every outgoing reply.
  4. ValidationMenzync Evaluate and audit the data before any model reaches a user.
A conceptual map of four ideas that run through my projects.

Selected work

Three projects, three ways of handling uncertainty.

All projects
01
  • Live

NextAct

Verify before you act.

A decision-support tool for the moment before you click, pay or share a code. It turns a suspicious message, link, document or payment request into claims, evidence, contradictions, unknowns, a risk level and the safest next step.

  • Next.js
  • TypeScript
  • pnpm monorepo
  • Zod
  • Drizzle
  • Vitest
  • Manifest V3
Read the case study : NextActVisit nextact.tech
What a result contains
  1. 1Claims
  2. 2Evidence
  3. 3Contradictions
  4. 4Unknowns
  5. 5Risk
  6. 6Safest next step

A result never says “safe”. The strongest one says no material risk was observed, and shows which facts were verified.

  • Case-submission and browser-extension APIs
  • Workspace-scoped database access
  • SSRF-controlled URL, DNS and TLS evidence collection
  • Adversarial API tests and GitHub Actions checks
  • Docker and Compose configuration
  • AI adapter interfaces with redaction and output validation
Two tracks, kept apart on purpose

Research

  • Roman Urdu / English text classifiers
  • Facial-expression experiments
  • Fusion design with abstention
  • Evaluation and data audits

Student prototype

  • Check-ins in English or Roman Urdu
  • Nothing saved by default
  • Consent-gated history
  • Export and delete

The dashed line is deliberate: no model is connected to the prototype until it has been evaluated.

  • Roman Urdu and English text classification: a TF-IDF + linear SVM baseline, XLM-RoBERTa, and a Gradio demo
  • A local FastAPI + SQLite prototype with consent-gated saved history, export and deletion
  • pytest and Vitest tests for account isolation and input validation
  • Evaluation tooling: macro-F1, per-class metrics, confusion matrices and calibration
02
  • Research
  • Prototype

Menzync

Emotion research and student check-ins.

Menzync explores affordable, private wellbeing support for university students in Pakistan. It runs as two separate tracks: emotion-recognition research, and a student check-in prototype.

Finalist, National Idea Bank IV 2026 · Co-founder & FYP developer

  • Python
  • FastAPI
  • SQLite
  • React
  • Vite
  • XLM-RoBERTa
  • scikit-learn
  • pytest
  • Vitest
Read the case study : Menzync
03
  • Prototype

EmoSense AI

An LLM backend that keeps working when the model doesn't.

A text-analysis app built for the CREATE X 2026 AI App Development Challenge. My part was the Express backend: Gemini calls constrained to structured JSON, input validation, logging, and fallbacks that keep the app usable when the model is unavailable.

  • TypeScript
  • Express
  • Gemini API
  • React 19
  • Tailwind CSS
Read the case study : EmoSense AIView repository
Every route follows the same path
  1. Validate input
  2. Call Gemini with a schema
  3. Or return a labelled fallback
POST/api/analyze
Emotion, sentiment, urgency, risk flags
POST/api/simulate-response
Reviews a drafted reply against the original
POST/api/persona-chat
Practice conversation with a persona
  • Three Express endpoints: text analysis, drafted-reply review, and persona practice chat
  • Gemini responses constrained by a JSON response schema
  • Request validation with clear 400 errors
  • Rule-based fallbacks when the API key is missing or a model call fails

How I work

Software that is honest about what it knows.

Four ideas run through most of what I build. None of them is a guarantee. Each is a specific engineering choice you can inspect in a real project.

  1. 01

    Evidence

    A result shows what was checked, where it came from, and what contradicts it. NextAct's results list unknowns next to findings.

    See it in NextAct
  2. 02

    Grounding

    Model output is held to a schema or to retrieved sources. EmoSense constrains Gemini to JSON; Aidesk answers from a company's documents with citations.

    See it in EmoSense AI
  3. 03

    Human approval

    Where a mistake is costly, software drafts and a person decides. ClientDesk sends nothing a freelancer hasn't approved.

    See it in ClientDesk
  4. 04

    Validation

    Tests, audits and evaluation come before claims. Menzync keeps its models out of the product until they've been measured.

    See it in Menzync

Experience

Where I've worked.

  1. AI Intern

    Aug 2026 – present

    FlyRank · Remote

    Audited my recurring work and sorted each task into delegate to AI, collaborate with AI, or keep human, then set up the tooling to match.

  2. Software Quality Engineering Intern

    Aug 2026

    Big Brains

  3. Change Makers Ambassador Intern

    May 2026 · six-week internship

    DataCrumbs

    Community outreach, AI literacy and student engagement.

  4. Transcription Typist

    Mar 2023 – Dec 2024

    TranscribeMe · Remote

    Transcribed and proofread audio and video to client formatting standards.

  5. Data Entry Clerk

    Jan – Mar 2022

    Silver Star Enterprises · Daska

    Entered and maintained invoice and customer records.

Recognition

Competitions & programmes.

  1. 3rd Place, All Pakistan Prompt Engineering Competition

    2026

    National-level competition (APPEC). Score 184.97, representing UMT Sialkot.

    Competition leaderboard row: rank 3, Muhammad Shahwaiz, University of Management and Technology Sialkot, score 184.97.
    From the competition leaderboard. Other participants' rows removed.

  2. Finalist, National Idea Bank IV

    2026

    Founders Edge / ASPIRE Pakistan pre-incubation programme, with Menzync.

  3. Top Performer, Generative AI Application Developer

    2025

    NCEAC-HEC intensive training.

Also

  • Merit certificate, essay writing, Government of the Punjab (A+)
  • Merit certificate, calligraphy, Government of the Punjab (A+)
  • Digital Pakistan Cybersecurity Hackathon, hands-on workshop, University of Sialkot, Aug 2026
  • Essay writing participation, Chief Minister Punjab Youth Festival
Training & courses (3)

Muhammad Shahwaiz

Daska · Sialkot · Pakistan

About

From transcripts to systems.

I'm from Daska and study Software Engineering at the University of Management and Technology in Sialkot. My route here wasn't direct: pre-medical studies first, then data entry, then two years of remote transcription, then software.

Those jobs were about getting messy information exactly right, and that habit carried over. The work I care about most now is AI software that shows where its answers came from, says what it couldn't check, and leaves the important decisions to a person.

I use AI coding tools heavily. I scope the problem, direct the build, review the code, and test it.

Education

  1. BS Software Engineering

    University of Management and Technology, Sialkot

    Sep 2023 – present · Final year

  2. Intermediate, Pre-Medical

    Superior College, Daska

    2019 – 2021 · Grade A+

  3. Matriculation, Science

    Govt. High School, Daska

    2016 – 2018 · Grade A+

Selected coursework

  • Deep Learning
  • Natural Language Processing
  • Information Security
  • Software Quality Engineering
  • Databases

Core tools

Python and TypeScript APIs (FastAPI, Express, Next.js route handlers), LLM integration, PostgreSQL and SQLite, pytest and Vitest, Docker.

In projects I've also used Celery and Redis, SQLAlchemy and Alembic, Prisma and Drizzle, Qdrant and FAISS, React and Tailwind. Coursework added deep learning, NLP, Wireshark and Android.

Languages

Urdu · Punjabi · English

Beyond code

  • Calligraphy

    Nastaleeq, Thuluth, Dewani and Suls. It earned a Punjab Government merit certificate, and it is the reason this site is built around a single pen line.

  • Chess

    I study classic games. It's a useful habit for designing systems too: always look for the reply you didn't expect.

  • Reading

    Whatever I can get through between coursework and building.

Contact

Building something that needs to be right? Let's talk.

For roles, collaborations or questions about any project here, email is the quickest way to reach me.

shahwaizarts@gmail.com