POMA AI

POMA turns your unstructured documents into accurate, retrievable context, so your AI, copilots, and agents all answer from a source you can trust.

[POMA AI @_POMA_AI_]

Most systems search images by squashing each one into a single vector, then wonder why retrieval keeps missing.
Dr. Alexander Kihm in TechRound on visual descriptions, and the small model doing the work.

Full interview → https://t.co/dTB2F7VDb1

[POMA AI @_POMA_AI_]

@spacedevin https://t.co/wncJ5xI5ra

[POMA AI @_POMA_AI_]

Most RAG pipelines are bad at tables for one reason: a fixed-size splitter cuts the table mid-row, and half your numbers lose their headers.
PrimeCut keeps tables as HTML. Rows and columns survive embedding, so the model gets a table.
Read more → https://t.co/lEdWhlNbu2

[POMA AI @_POMA_AI_]

Index4AI is live. Plug it into your agent or chatbot, ask a question about your documents, get an answer.

50+ file types, 80+ languages. The first 1,000 queries are free.

Try it yourself → https://t.co/uJTKNNbnZ4

[POMA AI @_POMA_AI_]

PrimeCut spends time reading the technical papers nobody else wants to read. It has now won an award for it :)
We took the AdMaCom 2026 Qualifier, run by INAM with MKS' Atotech, for turning scientific documents into machine-readable data!

Read more → https://t.co/TqhdJeyC1u

[POMA AI @_POMA_AI_]

Bring your worst PDF to booth T36 at the AI & Big Data Expo.

We run it through live: text, tables and images out, then chunks that follow the document's own structure instead of a character count.

How it works → https://t.co/euqB65LlLg

[POMA AI @_POMA_AI_]

Blueprints break most RAG pipelines. A fixed-size splitter cuts the drawing into character windows, so the legend lands in one chunk and what it explains lands in another.
PrimeCut cuts along the document's structure and keeps tables intact.
Check it out→https://t.co/HSs2Ij1jCc

[POMA AI @_POMA_AI_]

AI & Big Data Expo, Amsterdam, 19-20 October. Booth T36, next to the AI Innovation Pavilion.

Come talk clean context for your agents, messy tables turned into usable chunks, and RAG that runs on-premises or in the cloud.

30% off Gold Passes: SPONSOR30

[POMA AI @_POMA_AI_]

Boxing robots won the crowd at BIG BANG KI Festival. The most useful thing we heard had no robots in it.

Chunking is where most AI assistants break. PrimeCut cuts along the document's own structure: chapter, section, paragraph, table.

Free to try → https://t.co/dgEFmPiBEA

[POMA AI @_POMA_AI_]

POMA AI is at the AI & Big Data Expo in Amsterdam, 19-20 October. Booth T36, a few steps from the AI Innovation Pavilion.

Come talk RAG, context engineering, or token bills.

30% off Gold Passes: SPONSOR30

https://t.co/RSzEtCY0uH

[POMA AI @_POMA_AI_]

Our team - Alexander Nuss, Alexander Kihm, Jan Simon Raue - had an incredible time at BIG BANG KI FESTIVAL 2026 in Berlin!

Good conversations about applied AI, with people building it and people outside the field.

Thanks to everyone who stopped to talk!

[POMA AI @_POMA_AI_]

You can ingest 2,000 pages before we charge you a cent.

1,000 free when you sign up.
1,000 more when you review us on G2.

Console → https://t.co/HSs2Ij1jCc

[POMA AI @_POMA_AI_]

Sign up: 1,000 pages ingested free.
Leave a G2 review: another 1,000.
Say what you actually think. The pages don't depend on the rating.

Say your piece → https://t.co/VAhKRod5ty

[POMA AI @_POMA_AI_]

In most pipelines, the chart on page 7 becomes: [image]

Nothing in that figure is retrievable. PrimeCut describes it and drops that text into the chunkset. Now the chart is searchable.
€0.01 a page, tops.

Try it out for free → https://t.co/dgEFmPiBEA

[POMA AI @_POMA_AI_]

Most ingestion tools price every content type differently. PrimeCut has one ceiling: €0.01.

📃 1 document page
📹 1 minute of video or audio
🪙 1,000 tokens from a spreadsheet

First 1,000 pages free, across 50 filetypes.

Try it out → https://t.co/L0HUbOi8yj

[POMA AI @_POMA_AI_]

Your RAG pipeline has enough moving parts. PrimeCut adds exactly one: an API key.

Sign in to the POMA Console, generate it, and you're ingesting. Projects, roles and integrations are all in there too.

Start here → https://t.co/6iUFg6skPn

[POMA AI @_POMA_AI_]

Flat chunks give you a fact. Hierarchical chunksets tell you where the fact lives.

Every figure PrimeCut returns carries a root-to-leaf path to its exact spot in the document.

See it side by side: https://t.co/4IFRfRti5p

[POMA AI @_POMA_AI_]

The usual rule: the more complex the document, the worse the ingestion.
PrimeCut breaks the rule with a specialist parser for every format — Excel, CAD drawings, JPEGs, 50+ others.
Still max €0.01 a page.

Check it out → https://t.co/dgEFmPiBEA

[POMA AI @_POMA_AI_]

Most RAG pipelines treat a video file like a locked door. PrimeCut ingests audio and video — MP4, MOV, even WMV — and your first 3 hours are free.

Try it for free: https://t.co/HSs2Ij1jCc

[POMA AI @_POMA_AI_]

LLMs read like a bored student: sharp at the start, sharp at the end, hazy in the middle.

Dumping 200 pages on an agent doesn't give it understanding. Structure does.

Grill builds context around that attention curve instead of fighting it.

Read how we did it → https://t.co/KvoRDEwo3c