prasanna@builds:~$ project-detail

Inventory Sync (MFN + Marketplaces)

Comprehensive DAO governance platform enabling decentralized decision-making through smart contracts. Features quadratic voting, delegation mechanisms, time-locked proposals, and multi-signature treasury management.

Overview

Inventory was the heartbeat of Razor’s multi-channel commerce system — if it drifted, everything else broke.
Our products lived across Amazon MFN, Walmart, Shopify, Mercado Libre, and eBay.
Each platform had its own APIs, limits, and latency patterns, yet the customer expectation was binary:
if it says “in stock,” it must be true.

I built the synchronization layer that kept these worlds consistent.

The Challenge

When you sell across multiple marketplaces, every quantity update is a potential race condition.
ERP systems emit partial events, APIs fail mid-flight, vendors retry with new payloads —
and suddenly, 100 units in stock become 97 in one system and 110 in another.

At scale, this isn’t a bug. It’s entropy.
And entropy compounds silently until it breaks everything — from fulfillment to finance.

The Solution

The approach wasn’t to make it faster, but to make it consistent.
I designed a pipeline around truth propagation — every source event passed through three layers:

  1. Normalization: converting ERP deltas into structured channel updates.

  2. Debounce Logic: consolidating bursts into minimal writes, reducing API load and thrash.

  3. Reconciliation Engine: replaying and diffing every change against the source of truth.

The system favored determinism over throughput.
Each update carried a unique job signature — traceable, replayable, and immutable.
Failures went into a “dead-letter journal” that could auto-heal on retry or surface as actionable ops tasks.

When I built observability hooks, we could finally see our syncs — not just hope they worked.

The Impact

  • Maintained real-time stock accuracy across five major channels

  • Supported ~$10M GMV per month without oversells or stockouts

  • Reduced operational reconciliation overhead by 70%

  • Enabled “never out of sync” alerting within 15 seconds of drift

Reliability became measurable.
Ops teams stopped firefighting; they started trusting the system.

The Learning

The biggest lesson wasn’t about concurrency or APIs — it was about humility in systems design.
When you deal with thousands of moving parts, you can’t control the chaos — only contain it.
Building reliable infrastructure isn’t about clever code; it’s about discipline, visibility, and boring consistency.
The kind that outlasts its author.

Tech Stack

Python · Celery · Redis · PostgreSQL · Change Data Capture · S3 Logging · Grafana

Status

Confidential Internal Build
All source code and diagrams are private under Razor Group’s data compliance policy.

Runtime for AI Tools

A lightweight runtime that standardizes how AI agents define, execute, and trace tools, reliable by design, portable across frameworks, and auditable anywhere. Brings predictability, reusability, and observability to AI tool execution, the missing runtime between frameworks and production.

Sarah Chen

Tech Stacks:

© Prasanna

v24.10.2025