# Prism — Manufacturing Costing and Intelligence Platform

A six-category cost engine designed to mirror the factory's own cost-sheet methodology, with Gemini-powered PDF extraction and an approval-driven quotation workflow.

## Key facts

- **Domain:** MANUFACTURING
- **Status:** Built, deployed against a single mid-market manufacturer
- **Use when:** Mid-market manufacturers need to retire desktop quotation systems while preserving the operational discipline of their cost-sheet methodology.
- **Reference engagement:** A mid-market paint and plastics manufacturer

## The problem

Mid-market manufacturers run quotation on a maze of spreadsheets and a desktop system that grew up with the founder. The cost sheet methodology is precious operational discipline, and any replacement that loses it is dead on arrival. Off-the-shelf ERPs do not match the multi-layer costing reality (machines, components, products) that an actual factory runs on.

## The approach

Prism is a multi-layer costing platform that mirrors how the factory actually thinks: machines with shift tariffs, components with their own material BOMs, finished products assembled from components plus overheads. The cost engine breaks every quotation into the same six categories the factory uses on its physical sheet, and is calibrated against the customer's existing methodology so the numbers it produces match the discipline already in place.

## The architecture

Built as a modular monolith on Next.js 14 (App Router), deployed to Firebase App Hosting (Cloud Run). PostgreSQL is the primary data store; Firestore handles real-time approval notifications. The cost snapshot per quotation line stores the full category breakdown as JSONB so historical quotations do not drift when machine tariffs or item prices change later. AI-powered PDF extraction uses Vertex AI Gemini 3 Flash to parse OEM requirement PDFs, then `pg_trgm` fuzzy matching against the catalogue with confidence classification (auto, review, unmatched). The quotation workspace converges three input methods (manual product add, AI PDF extraction, OEM template) into the same cost engine. Approval workflows fire Firestore notifications in real time.

## The outcome

A modern web quotation platform that mid-market leadership trusts because it produces the same numbers the cost sheet produces. AI lifts the data-entry burden of OEM requirement PDFs without anyone having to trust an LLM with the actual costing logic. The approval workflow brings real-time visibility to a process that used to live in a manager's inbox.

## ATI shape

- Predictive intelligence — Cost prediction across multi-layer BOM, with the model calibrated to physical cost sheets
- Agentic execution — Gemini-powered PDF extraction with `pg_trgm` fuzzy matching for catalogue resolution
- Secure data — Private PostgreSQL on the customer's cloud; no data leaves their perimeter
- Actionable outcomes — Quotation decisions with an approval workflow and real-time notifications

## Tech stack

- Next.js 14
- TypeScript
- PostgreSQL 15
- Firebase Auth
- Firestore
- Vertex AI Gemini 3 Flash
- pg_trgm
- @react-pdf/renderer

## Impact

- Multi-layer — BOM costing across machines, components, and finished products
- 6 — Cost categories matching the physical sheet methodology
- AI-extracted — OEM requirement PDFs into catalogue matches
- Auditable — Cost-sheet snapshots per quotation line

## Related

- [All accelerators](https://levent.ai/accelerators/)
- [Next: Pitlane](https://levent.ai/accelerators/pitlane/)

Metrics on this page are estimated and expected improvements describing the design intent of the accelerator. Real-client delivered metrics stay in private decks; see https://levent.ai/ai-content-policy/ for the abstraction policy.

---

**Canonical URL:** https://levent.ai/accelerators/prism/