All projects

Backend / B2B marketplace / 2026

SurplusLink AI

MVP · Simulated checkout

A B2B marketplace MVP connecting suppliers of unsold packaged bakery inventory in Pakistan with business buyers before products expire.

My role

Full-stack development · Backend & AI integration

Skills & technologies

  • React
  • FastAPI
  • Python
  • PostgreSQL
  • GroqCloud
  • Docker

The Problem

What needs to change

Suppliers of unsold packaged bakery products need to find business buyers before stock expires. Buyers need enough information about quantity, location, price, and shelf life to decide whether surplus stock fits their needs.

The Solution

How SurplusLink AI helps

SurplusLink AI connects listings and buyer requirements through explicit matching rules. Suppliers can review AI-assisted listing extraction, buyers can make offers, and stock reservations support the collection workflow. Checkout remains simulated in this MVP.

01 / Contribution

What I worked on

Built listing and buyer workflows, integrated reviewable AI extraction, and implemented matching, stock reservations, and simulated transaction reporting.

02 / Capabilities

What the project includes

  • Surplus listings, buyer requirements, offers, stock reservations, and collection tracking.
  • GroqCloud LLM integration with structured JSON extraction, Pydantic validation, and user review.
  • Bidirectional rule-based matching by category, location, quantity, budget, and shelf life.
  • Simulated checkout and transaction reporting; payments are not real.

03 / Architecture

System overview

  1. 01React listing & buyer workflows
  2. 02FastAPI routes / Pydantic contracts
  3. 03Service rules / SQLAlchemy models
  4. 04PostgreSQL transactions & reservations
  5. 05GroqCloud extraction / user review

Responsibility map · Read alongside the implementation status above.

04 / Decisions

Engineering notes

Review before persistence

AI extraction returns structured suggestions for user review. It does not publish inventory or replace deterministic validation.

Transactional reservations

Inventory reservation and offer acceptance are handled together, using PostgreSQL transactions and row locks to address competing requests.

Deterministic matching

Matching uses explicit marketplace criteria. LLM extraction supports data entry rather than deciding whether an offer is valid.

05 / Validation

Validation & scope

The backend repository documents an end-to-end simulated demonstration and tests for permissions, extraction failures, reservation races, payment retries, cancellation, and collection. Those repository tests have not been rerun as part of this portfolio work.

06 / References

Source & technical evidence

Next project

Job Junction