- Entry date
- 1 August 2025
- Category
- Tooling
- Lead over the world
- the honesty, not the lead
- Access
- 🔓 Public
38% complete, frontend never wired to the backend, running on mock data — stated plainly, because an honest failure is worth more to this archive than a polished half-truth.
- 38%
- complete, per the project's own honest self-assessment
- 0
- real requests flowing frontend-to-backend at the time of the audit
- 4+
- parallel subsystems attempted at once: pgvector, MeiliSearch, MinIO, Redis-backed queue
Honest evaluation
The core request-to-response loop was never closed — the frontend ran on mock data, never wired to the backend, at a measured 38% complete.
What would prove or disprove it further
What would prove the alternative ordering: rebuilding the same scope but closing the core request-to-response loop first, before layering pgvector, MeiliSearch, MinIO, and the queue on top — which is exactly the lesson this report draws and the pattern a later, more disciplined project (linked above) followed instead.
The evidence — full reasoning behind the verdict
Verdict: disproven. The implicit hypothesis behind building four-plus subsystems in parallel — that a vector database, a dedicated search engine, object storage, and a task queue could all be wired up simultaneously and still arrive at a working system — did not hold. Measured against its own bar, the project stalled at 38% complete with the frontend never connected to the backend at all; what was on screen was mock data, not a working pipeline.
This isn't ambiguous or partial in the way "partly-proven" would suggest — the one thing a tender-management system has to do, take a real request and answer it end-to-end, never happened. The mechanism is the sequencing choice itself: breadth (four subsystems at once) was prioritised over depth (one closed loop), and the loop that would have proven the system works was never closed as a result.
Most of this archive is honest about small gaps in otherwise-working systems. This one is different: it's an honest account of a project that did not close its core loop, told plainly rather than smoothed over.
What it was
TenderMS — an AI-powered tender-management system intended to handle document processing, AI analysis, proposal generation, and workflow automation for tender bids.
What we built
ResearchAn ambitious stack: Next.js 14 and FastAPI, PostgreSQL with pgvector, Redis, MinIO for object storage, Docker, ChromaDB, MeiliSearch, and an LLM API — four-plus significant subsystems, all being built at once. The project's own self-assessment, stated here without softening, put it at 38% complete, with the frontend disconnected from the backend and running on mock data.
What we learned — including the honest negative
This is the whole point of the entry, and it belongs in "What we learned," not hidden in a footnote: what's on screen for TenderMS was never the real pipeline. The frontend and backend were never connected. The demo data was mock data.
That's a specific, transferable lesson, not a vague "it was hard": breadth before depth, applied to infrastructure, delays the one integration point that actually proves the system works at all.
Where it went / status
Left at 38% complete, frontend and backend disconnected, per the project's own honest measurement. Kept in the archive as a genuine, useful "here is what an overscoped integration looks like" entry — valuable precisely because it is not a success story, and because the ordering lesson it produced shaped how later, more disciplined platform work in the portfolio was sequenced.
What is still open — kept visible
The honest edges, next to the wins. This is what turns 🔬 into 🟢 — honestly.
- The core request-to-response loop was never closed end-to-end — what's on screen is mock data, not the real pipeline. This is the whole point of the report; it is not buried here.
- Too many integrations were attempted simultaneously; the honest lesson is that the core loop should have worked first, with the rest layered on after.
Where this connects
Sources
- vocabotics project audit — TenderMS (38% complete, frontend/backend disconnected), Aug 2025vocabotics internal project history · as of August 2025