AI Enterprise SolutionsGPT & Claude

SortSight AI — Intelligent Parcel Inspection & Sorting

Logistics companies and warehouse operators often rely on manual parcel label inspection and sorting. This process is time-consuming, prone to human error, and difficult to scale. SortSight AI uses Gemini Vision to extract tracking numbers and delivery destinations from parcel labels, identify missing or inconsistent information, and recommend sorting lanes or manual review. The goal is to reduce repetitive inspection work and improve sorting accuracy.

Open the demo GitHub

Preliminary score

52.5/ 100Earlyaverage of 2 judges

Value for the user15.5 / 25
A specific problem and a clear outcome from solving it.
Prototype and use of AI17.5 / 30
A working core scenario, and what AI actually contributes to the result.
Quality testing9.0 / 20
Tests, examples of failures, and comparison with the current approach.
Feasibility6.5 / 15
Data requirements, running costs and a clear next step.
Originality4.0 / 10
A meaningful difference from common solutions.

What the judges wrote

GPT-6 AstraPreliminary59/100
Strengths
Your screenshots show inspection results, recorded history, and a sorting-rule test, providing concrete evidence beyond a concept deck. Your separation of Gemini extraction from explicit routing rules is sensible, and the unsupported-destination example demonstrates a useful manual-review fallback. Your disclosure of AI failures, deployment connectivity problems, and demonstration data makes the prototype's limitations clear.
To improve
Show a supported parcel image reaching its assigned lane end to end, and resolve the acknowledged frontend–backend connectivity issue. Reconcile the sample output's name and street details with the displayed label, where those details are not visible despite an output confidence of 1. Benchmark labeled clean, damaged, and ambiguous examples against manual inspection or OCR, reporting extraction accuracy, misroutes, review rate, and processing time. Add per-parcel running costs, label-data privacy handling, and a small warehouse pilot with a measurable adoption target.
Claude Opus 5.5Preliminary46/100
Strengths
Your pipeline is genuinely wired together: the screenshots show real Gemini Vision extractions (tracking number, full address, confidence) feeding a rule-based engine that returns SORT or MANUAL_REVIEW with a reason. Defaulting unsupported destinations to manual review instead of guessing a lane is a sensible human-in-the-loop design. Your write-up and README are honest about limitations, including the broken public deployment, demo dashboard data and local-storage history, and that candour is valuable.
To improve
Build a labelled test set of 30–50 label images, including clean, damaged, handwritten and partially obscured labels, and report extraction accuracy per field, along with the specific labels where Gemini failed. Compare against today's process: manual keying time per parcel, or a standard barcode scanner plus OCR, because the track asks for a metric that justifies adoption and none is named. Your deck's own demo sends both real packages to review, so show a case where the AI clearly saves work. Add cost per inspected label and latency so a warehouse can judge feasibility at conveyor speeds.

The team

Aysel Musayeva
developer
Zahra Rufullayeva
designer
Asma Mutallimli
product
Nurana AliyevaLead
data

Disclosure

Required by the rules. Judges verify it against the repository.

Models: Google Gemini Vision for parcel label image analysis and structured information extraction.
Data: User-uploaded parcel label images and manually created test examples. No proprietary logistics dataset was used.
Components: Next.js, React, FastAPI, Python, Google Gemini API, rule-based sorting engine, Vercel and Render.
Built during the event: Confirmed

Quality testing

We tested parcel label extraction, tracking number and destination recognition, sorting decisions, and dashboard inspection history. Backend automated tests passed (11 tests). We verified that unsupported destinations are routed to manual review rather than assigned an incorrect lane. During development, we encountered transient AI analysis failures and an inspection history integration issue, which was resolved locally. The public deployment currently has a frontend-to-backend connectivity issue that remains to be fixed. Testing is prototype-level; large-scale accuracy and performance benchmarks have not yet been conducted.

Worth 20 of the 100 points.

Running it

Open the public demo and navigate to AI Inspection. Upload a parcel label image and select Inspect Package. The backend uses Gemini Vision to extract label information and a rule-based sorting engine to determine whether the parcel should be sorted or sent for manual review. Results can be viewed in the dashboard inspection history. Source code and local setup instructions are available in the GitHub repository. The public deployment may currently have an API connectivity issue; the core workflow was verified locally.

Project ownership

The team keeps the rights to this project. Judges were given access for verification only.