Muhammad Shahwaiz

Work

Everything I've built, labelled honestly.

Each project carries a stage label: live, prototype, research, in progress or academic. I use AI coding tools heavily. I scope the problem, direct the build, review the code, and test it.

Product prototypes

Working builds that run locally. Integrations that haven't been tested against the real vendor are described as such.

  • ClientDesk

    Approval-first client operations for freelancers on WhatsApp.

    • Prototype

    Triages inbound client messages and drafts replies that a person must approve. Client facts are stored only with a verbatim quote checked against the source message. Adds scope briefs, scope-change detection, proposals, follow-ups and invoices. WhatsApp runs through a local simulated provider; the real integrations have not been run against the vendors.

    • FastAPI
    • PostgreSQL (RLS)
    • Celery
    • Redis
    • Next.js
    • TanStack Query
  • Aidesk

    Document-grounded customer support.

    • Prototype

    Answers customer questions from a company's own documents with citations. Local MiniLM embeddings and a per-tenant FAISS index, a confidence gate before the model is called, an extractive fallback when no API key is set, and a human escalation queue.

    • FastAPI
    • FAISS
    • Sentence-Transformers
    • Gemini
    • Next.js
  • DocFlow AI

    Document processing with a human review step.

    • Prototype

    A six-stage pipeline (preprocess, classify, extract, validate, route, export) where each stage is a retryable background task. Versioned extraction schemas, decimal-exact validation, a review workspace, and a swappable model provider.

    • FastAPI
    • SQLAlchemy
    • PostgreSQL
    • Celery
    • Redis
    • Next.js
    • Radix UI
  • Beacon

    Knowledge search with cited answers.

    • Prototype

    Indexes uploaded company documents and answers questions with a synthesised, cited answer instead of a list of links. Tenant isolation through Postgres row-level security, Qdrant for retrieval, Claude for synthesis and reranking.

    • FastAPI
    • PostgreSQL (RLS)
    • Qdrant
    • Celery
    • Claude API
    • Next.js
  • Vantra

    Business management for small teams.

    • Prototype

    A multi-tenant workspace for customers, projects, Kanban tasks and invoicing. Invoice amounts are stored in integer minor units so totals round exactly. Includes an assistant that answers questions about the workspace's own data.

    Vantra marketing homepage with the headline 'Run your entire business from one intelligent workspace.'
    • Next.js
    • React
    • Tailwind CSS
    • Prisma
    • PostgreSQL
  • AI Sales & Lead Automation

    Background lead scoring and outreach drafts.

    • Prototype

    Clients describe their target customer and import leads. Claude scores each lead and drafts outreach in background jobs, so the API never waits on a model call. Includes Stripe webhook idempotency, rate limiting and request tracing.

    • FastAPI
    • Celery
    • Redis
    • PostgreSQL
    • Claude API
    • Stripe
    • Docker
  • Competitive Intelligence System

    Structured competitor reports from grounded search.

    • Prototype

    Gemini with Google Search grounding researches recent competitor activity and returns a structured report and PDF. Schema fallback, exponential backoff and a concurrency cap keep it inside free-tier rate limits.

    • FastAPI
    • Gemini
    • Supabase
    • ReportLab
    • Next.js
  • ORVIA

    Trade operations platform.

    • In progress

    An early-stage platform for exporters, covering buyers, products, quotes, orders, documents and export readiness.

    • Next.js
    • React
    • Prisma
    • TypeScript
    • Zod
    • Vitest

Coursework & competitions

University projects and competition entries.

  • CreateX text analysis app

    Sentiment and zero-shot categorisation.

    • Academic

    A Streamlit app using DistilBERT for sentiment and BART-Large-MNLI for zero-shot categories, with bulk CSV processing and a Plotly dashboard. Built for CreateX 2026 at UMT Sialkot.

    • Python
    • Streamlit
    • Hugging Face Transformers
    • SQLite
    • Plotly
  • Netflix testing study

    Software Quality Engineering course project.

    • Academic

    A test plan, manual and automated test cases, and bug reports for the Netflix website and app, written with Saud Baseer as coursework. Not affiliated with Netflix.

    • Test planning
    • Manual testing
    • Bug reporting
  • Deep learning coursework

    Proposal and experiments.

    • Academic

    A BiLSTM + 1D-CNN term-project proposal, and a transformer experiment on the FER2013 facial-expression dataset.

    • Python
    • PyTorch
    • TensorFlow
  • Network security labs

    Packet analysis coursework.

    • Academic

    Wireshark capture and filtering labs on TCP, HTTP and DNS traffic, plus a cybersecurity audit course project.

    • Wireshark
    • Networking
  • Student grade manager

    C++ console application.

    • Academic

    Manages student records and GPA calculation with object-oriented design and file storage.

    • C++
    • OOP
  • Android task manager

    Offline productivity app.

    • Academic

    Task categories, priorities and reminders with offline SQLite storage.

    • Java
    • Android Studio
    • SQLite