Flagship project, BSc final-year project

Intelli-Factory

A multi-objective optimization platform for supply-chain matching

A B2B2C platform that matches customer requests with manufacturer-logistics pairs across the “supply-chain trilemma”: cost, delivery time, and reliability. Four user roles, a nine-state request lifecycle enforced by explicit state machines, three-party contract signing, and an admin UI for comparing optimization strategies live.

Optimization engine

NSGA-II-style genetic algorithm built on DEAP (population 100, 80 generations) plus a fast weighted strategy and a greedy baseline, producing Pareto-optimal sets with knee-point selection.

Production platform

FastAPI + PostgreSQL (~25 Prisma models) behind a Next.js 14 App Router frontend; deployed as a three-tier system on Vercel, Render, and Aiven with Docker Compose for local dev.

Security & testing

OWASP-aligned: Argon2id hashing, CSPRNG server-side sessions, rate-limited login. 51 automated pytest unit and integration tests, TDD on the engine.

Pareto front, Deep GA

  • Non-dominated solution
  • Knee point (selected)
  • Greedy baseline
Pareto front computed by the Deep GA: cost in thousands of KZT against delivery time in days, with the greedy baseline for comparison.
SolutionDelivery time (days)Cost (thousand KZT)
Solution 13.260.2
Solution 23.757.4
Solution 34.154.6
Solution 4, knee point, selected4.6751.6
Solution 55.445.8
Solution 66.139.9
Solution 7733.2
Solution 87.826.4
Solution 98.422.9
Greedy baseline8.0221.3
The Deep GA (NSGA-II-style) Pareto front, drawn as the engine computes it. Each mark is a non-dominated manufacturer-logistics pairing trading cost against delivery time. Benchmark figures on this page come from the real 3,600-run evaluation.

Verified benchmark

+17.5%
composite fitness vs greedy baseline
41.8%
faster delivery (8.02 → 4.67 days)
+8.1%
reliability score (0.824 → 0.891)
0.069 s
avg Deep GA response time
100%
feasibility across 120 scenarios
3,600
benchmark evaluations (120 × 30 seeds)

Verified benchmark: 120 synthetic scenarios × 30 random seeds, run on the production engine code.

Production architecture

  1. Frontend

    Next.js 14 · TypeScript · Tailwind

    Vercel

  2. API

    FastAPI · Python 3.12 · DEAP

    Render

  3. Database

    PostgreSQL · Prisma

    Aiven

Docker Compose for local development, 51 automated pytest tests, Ruff and ESLint, Brevo SMTP email verification, and a state-machine-enforced request lifecycle.

Live demo(opens in a new tab)API docs (Swagger)(opens in a new tab)Source & README(opens in a new tab)

Deployed and running right now. The free-tier API cold-starts in roughly 50 seconds on the first request.

Standard of evidence

Not promises. Measurements.

Every metric on this page comes from a 3,600-run benchmark on production code, and it is reproducible with a single command.

benchmark_evaluation.pyRerun it yourself.

Keep exploring