Tayyib Trace

Inspiration

I started out asking why ethical consumption is so hard, and the answer kept pointing upstream. Most companies can't actually see their own supply chains. The normal way to fix that is hiring a supply chain consulting firm, which costs tens of thousands of dollars and takes weeks. Small and mid size manufacturers just can't afford it, so they operate blind until a tariff or an export ban or a supplier collapse turns some dependency they never knew about into an emergency. I wanted to build that visibility for the companies consultants ignore.

What it does

You describe what your company makes, and if you want, paste in whatever you know about your suppliers. It could be a full bill of materials or just a note like "batteries from a supplier in Poland." Tayyib Trace maps out your component sourcing in about a minute. Each component gets scored on four kinds of risk: geopolitical exposure, labor and ESG issues, environmental impact, and supply concentration, meaning the single source dependencies that sink companies. There's a dashboard with the headline numbers, every component expands to show detail and alternative sourcing regions, and one click generates a downloadable executive report with a summary, the key exposures, and prioritized actions. It's written for leadership, not the logistics team.

How I built it

React on the frontend, with the Claude API doing the actual analysis. The trick that made it work was treating the prompt like a typed contract: explicit JSON schema, fixed values for risk levels, word limits on every field. That made the output reliable enough to drive a dashboard. Whatever supplier notes the user pastes in get injected into the prompt, so the AI anchors to real data where it exists and fills the gaps with typical industry sourcing where it doesn't. The report is a separate second call that takes the structured analysis and writes it up as prose. Splitting it into two calls keeps everything fast and under token limits. The whole UI is styled like a customs manifest, with tag shaped components, dashed freight routes, and a rubber stamp exposure grade.

Challenges I ran into

Getting an LLM to return strictly parseable JSON every single time took a lot of iteration. I ended up with schemas, word limits per field, and defensive parsing that strips markdown fences. The harder problem was messy input. Real companies almost never have a clean bill of materials, so the prompts had to blend fragments of user data with inferred industry sourcing without dressing up guesses as facts. That forced my biggest product decision: every assessment is labeled as an AI first pass that should be validated with supplier audits. A risk tool that overclaims would create exactly the blind spots it's supposed to fix.

Accomplishments that I'm proud of

It works with whatever data a company actually has, whether that's nothing but a product name or a detailed supplier list. To me that's the difference between a demo and something a real ops manager could use on day one. The part I'm proudest of is the pipeline from analysis to deliverable: two clicks gets you from a risk assessment to a downloadable executive report, which compresses what's normally a multi week engagement into a lunch break. I'm also glad I resisted the urge to overclaim. The honesty labeling is a feature, not fine print.

What I learned

Prompt design is basically API design. Once I treated prompts as typed contracts, reliability stopped being a problem. I also learned that the hard part of B2B AI tools isn't the analysis itself, it's meeting users where their data actually is, which is messy, partial, and half remembered. And pivoting taught me something: my first version told consumers what to buy. Sitting with the problem longer showed me that businesses have the same blindness, higher stakes, and an actual budget to fix it.

What's next for Tayyib Trace

Grounding the assessments in public data like trade databases, sanction lists, and certification registries, so the results move from informed estimates to sourced findings. Continuous monitoring, so a new export restriction pings you about the component it affects instead of waiting for you to ask. Letting suppliers verify and correct their own data, which turns the assessment into a shared source of truth. And better deliverables: PDF and slide exports, scenario comparisons like "what if we moved frame production to Vietnam," and integrations with the ERP systems where bills of materials actually live.

Built With

• react