Cisco Chemicals - B2B Manufacturing Platform
B2B chemical procurement portal with intelligent inventory search and automated RFQ workflows.
Executive Summary
The Challenge
Manual quote-to-order bottlenecks, multi-tier RFQ approvals, and AI hallucination risks on complex industrial chemical specifications.
The Solution
Headless PayloadCMS 3.0 on Neon Postgres, context-grounded Gemini 2.5 Flash query pipeline, and React-PDF compilation.
Measurable Impact
Automated RFQ workflows, grounded catalog responses constrained to live inventory records, and fully integrated Stripe-to-Resend billing lifecycles.
Technologies
Key Decisions
Why PayloadCMS over traditional headless CMS?
PayloadCMS runs directly inside Next.js without a separate server. Its native TypeScript schema generation and collection-level hooks gave type-safe RBAC and direct database access with zero API latency.
Why context-grounded prompting over fine-tuning or vector embeddings?
Chemical catalog inventory and stock statuses fluctuate in real-time. Fine-tuning is static and costly to re-train, while vector embeddings can lose exact SKU precision. Injecting structured catalog data directly into Gemini's system prompt with strict schema constraints guarantees exact alignment with live database stock without vector database infrastructure overhead.
Why React PDF?
We needed pixel-perfect, automated invoice generation. React PDF allowed us to build invoices using the same component logic we use for the web.
Retrospective
Hardest Part
Orchestrating the post-purchase webhook lifecycle where Stripe confirms payment, triggers Payload hooks, generates a PDF, and emails the client via Resend.
Biggest Insight
Serverless databases (Neon) are incredible for scaling, but managing connection pooling in serverless functions is critical to prevent cold start timeouts.
Do Differently
I would introduce a queueing system like Upstash/Redis for the email processing to ensure the checkout UI resolves instantly regardless of third-party API delays.
Key Highlights
B2B RFQ System
Complete 'Request for Quote' workflow allowing admins to approve custom prices & quantities.
AI Inventory Assistant
Context-grounded assistant using Gemini 2.5 Flash that answers queries strictly based on verified inventory.
Dynamic Invoicing
Automated generation of downloadable PDF invoices with unique transaction IDs post-payment.
Admin Analytics
Visual dashboard tracking monthly sales, quote conversion rates, and catalog inventory status.
Next Project
envspect — Zero-Config .env & Secret Audit CLIInterested in discussing this project or exploring collaboration? Let's talk