Pipely

Landing Page
Standard, n < 80% (Needs to be scheduled based on law)
Standard, n > 80% (Informational)
Datasets + Alignment Jobs (Both ML/No ML)
Matches
Visualization
Auditing (Logging)
Login Page

Inspiration

Pipeline integrity management is a critical safety concern in the oil and gas industry. Inline Inspection (ILI) tools generate massive datasets that need to be aligned across multiple inspection runs to track corrosion growth, detect anomalies, and predict failures. Currently, engineers spend countless hours manually aligning this data in spreadsheets — a tedious, error-prone process that can have serious safety consequences if done incorrectly.

We built Pipely to automate this entire workflow with AI, turning what used to take days into minutes.

What it does

Pipely is a full-stack platform that automates ILI data alignment through a 7-stage pipeline:

Ingest - Drag-and-drop upload of ILI inspection datasets (CSV/Excel)
Normalize - Standardizes column names, units, and feature types using Gemini AI
Anchor - Identifies weld-to-weld reference points across inspection runs
Correct - Applies clock-position and distance corrections
Match - Pairs anomalies across runs using deterministic + ML similarity scoring
Score - Enriches matches with industry standards (ASME B31.8S, API 1163, NACE SP0502)
Export - Generates audit-ready reports with PHMSA compliance records

The platform also features:

AI-powered canonicalization via Google Gemini for inconsistent field names
XGBoost ML sidecar for enhanced similarity matching (80/20 deterministic/ML blend)
Interactive visualizations for aligned pipeline data
Standards assessment with severity classifications and growth analysis

How we built it

Frontend - Next.js 16 with React 19, styled with Tailwind CSS 4. The landing page features a GSAP-animated mountain landscape with drifting clouds and falling snow, using Instrument Serif typography for an elegant, modern feel.

Backend - Next.js 14 API routes with Mongoose ODM connecting to MongoDB Atlas. The 7-stage pipeline is orchestrated through numbered TypeScript modules with pure utility functions. Authentication via NextAuth with JWT sessions.

ML Sidecar - Python FastAPI service running XGBoost for similarity scoring and DBSCAN clustering, with a no-op fallback for graceful degradation.

Architecture - Monorepo with proxy-based communication. The frontend proxies API calls to the backend via Next.js rewrites, keeping both apps independently deployable.

Challenges we ran into

Standards compliance - Implementing ASME B31.8S severity calculations, API 1163 tool qualification logic, and NACE SP0502 corrosion growth classification required deep domain research
Data normalization - ILI datasets come in wildly inconsistent formats; we used Gemini AI to canonicalize field names but had to carefully guard against hallucinated mappings
ML blending - Balancing deterministic matching with ML predictions required an 80/20 cap to ensure the ML sidecar remains advisory, never overriding proven matching logic
Two Next.js versions — Running Next.js 14 (backend) and Next.js 16 (frontend) in the same monorepo required careful proxy configuration and port management

Accomplishments that we're proud of

Full 7-stage pipeline that processes real ILI inspection data end-to-end
Industry-standard compliance built in (ASME, API, NACE, PHMSA)
Beautiful animated landing page that makes pipeline data feel approachable
Graceful degradation - the system works without Redis, without the ML sidecar, and without Gemini AI