Paying three AI vendors to parse our own docs: a Reddit quest for one self-hosted stack
Sad-Razzmatazz-7657 · reddit · 2026-09-16
An engineer at one of Andrew Ng's companies vents on Reddit that a single document pipeline bills OpenAI (embeddings), Cohere (reranking) and Textract (OCR) separately — three SDKs, three rate limits, and data leaving their cloud three times. Self-hosting options surveyed:
- Piece together: TEI + separate reranker + docling for OCR; works but means managing three deployments
- Infinity: covers embeddings + reranking on one server, no OCR
- SIE: the only option found bundling all three behind one API; Superlinked cites docling at $0.397 per 1,000 pages vs $1.50 for managed services, though OCR quality numbers are unpublished
Data residency, not cost, is the deciding factor since contracts can't leave their cloud. The poster asks for real-world OCR-on-contracts experience.
More from Infra
- MediaTek's 2nm Dimensity 9600Pro runs 30B MoE models on-device with rebuilt dual NPUs — 量子位 · 2026-09-16
- Transformers models now run natively in vLLM with no port required — pcuenq · 2026-09-16
- TSMC builds 20 fabs yet can't meet AI demand as labor shortage slows expansion — emmanuelvivier · 2026-09-16
- SemiAnalysis: Nvidia Vera Rubin NVL72 Delivers Up to 30x Higher Throughput per MW Than Blackwell for Agentic Inference — emmanuelvivier · 2026-09-16
- Cacheon Miners Push MiniMax M3 to 2,337.9 tok/s, +32.2% Over SGLang — Modeled as ~61% More Profit per Chip-Hour — JosephJacks_ · 2026-09-16
- Nadella: a 400-500MW data center grew a rural town's tax revenue 12x — rohanpaul_ai · 2026-09-16