Access unstructured OSS data with object table

Optional parameters

Query an object table

After the initial refresh, query an object table like any standard table:

SELECT * FROM <object_table_name>;

Object tables have the following fixed columns:

To view the DDL or storage properties of an object table, use either of the following:

-- View DDL
SELECT hg_dump_script('[schema_name.]<object_table_name>');

-- View storage properties
SELECT * FROM hologres.hg_table_properties WHERE table_name = '<object_table_name>';

REFRESH OBJECT TABLE [schema_name.]<table_name>;

-- Rename
ALTER OBJECT TABLE [IF EXISTS] <table_name> RENAME TO <new_name>;

-- Move to a different schema
ALTER OBJECT TABLE [IF EXISTS] [schema_name.]<table_name> SET SCHEMA <new_schema>;

-- Moves to the recycle bin if it is enabled; otherwise, permanently deletes
DROP OBJECT TABLE [IF EXISTS] <table_name>;

-- Permanently deletes without moving to the recycle bin
DROP OBJECT TABLE [IF EXISTS] <table_name> FORCE;

Create a table to store embedded PDF data.

Create an internal Hologres table to store embedded PDF data. Build a vector index on this table for vector search.

-- Create embedding table
CREATE  TABLE public.embed_pdf_bs (
    object_uri text NOT NULL,
    etag text NOT NULL,
    chunk text NOT NULL,
    embedding_vector float4[] CHECK(array_ndims(embedding_vector) = 1 AND array_length(embedding_vector, 1) =1024)
) WITH (
bitmap_columns = 'object_uri,etag,chunk',
clustering_key = 'object_uri,etag',
distribution_key = 'object_uri',
vectors = '{
      "embedding_vector": {
        "algorithm": "HGraph",
        "distance_method": "Cosine",
        "builder_params": {
        "base_quantization_type": "sq8_uniform",
        "max_degree": 64,
        "ef_construction": 400,
        "precise_quantization_type": "fp32",
        "use_reorder": true
        }
      }}'
      );

Call AI Function to embed and chunk PDFs, then write results to the table.

-- Convert PDFs to documents and embed
insert into public.embed_pdf_bs 
  WITH parsed_doc AS (
    SELECT object_uri, etag, ai_parse_document(file, 'auto', 'markdown') AS doc 
    FROM pdf_bs_challenge_financial_14b_dataset limit 1), 
  chunked_doc AS (
    SELECT object_uri, etag, unnest(ai_chunk(doc, 300, 50)) AS chunk 
    FROM parsed_doc )
  SELECT object_uri, etag, chunk, ai_embed(chunk) AS embedding_vector 
  FROM chunked_doc
  WHERE chunk IS NOT NULL AND length(chunk) > 0;

Search and infer PDF content.

You can assess whether a company’s future outlook is optimistic or pessimistic based on financial trends in prospectuses. This helps guide investment decisions. Example SQL:

WITH
  -- Embed question and retrieve
  embedding_recall AS (
    SELECT chunk FROM embed_pdf_bs
    ORDER BY
    approx_cosine_distance(
      embedding_vector,
      ai_embed('In the reporting period, what were the year-on-year growth rates for revenue and net profit of Hunan Goke Microelectronics Co., Ltd. in 2014, 2015, and 2016?')
    ) DESC
    LIMIT 20
  ),
  -- Concatenate top chunks
  concat_top_chunks AS (
    SELECT string_agg(chunk, E'\n\n----\n\n') AS merged_chunks FROM embedding_recall
  )
-- Generate final answer
SELECT
  ai_gen('Analyze whether the following financial trend is optimistic or pessimistic, and explain why: ' ||
         ai_gen('In the reporting period, what were the year-on-year growth rates for revenue and net profit of Hunan Goke Microelectronics Co., Ltd. in 2014, 2015, and 2016? '  || merged_chunks))
FROM concat_top_chunks;

The query returns:

ai_gen
----
"From the provided financial data and growth analysis, Hunan Goke Microelectronics Co., Ltd. showed an overall **optimistic** performance trend from 2014 to 2016, for the following reasons:

### 1. **Revenue grew consistently**
- Revenue increased by 112.94% in 2014, 103.00% in 2015, and 33.22% in 2016. This indicates steady business expansion and rising market demand.
- Growth was driven by mass production and sales of new products (e.g., smart surveillance chips and GK-series chips) and partnerships with well-known distributors, highlighting strengths in product innovation and market outreach.

### 2. **Net profit rose significantly**
- Net profit surged by 467.53% in 2014, 8.21% in 2015, and 33.07% in 2016. Although growth slowed in 2015, it remained positive.
- In 2016, higher-margin products (e.g., solid-state storage chips) contributed more to revenue, while sales expenses declined. This reflects successful optimization of product mix and cost control.

### 3. **Product mix improved and innovation advanced**
- In 2014, the company benefited from the opening of the Live Satellite market. In 2015, smart surveillance chips entered mass production. In 2016, GK-series chips were adopted by major distributors. This shows the company’s ability to align with market needs and drive growth through innovation.
- Higher-margin products made up a larger share of sales, boosting overall profitability.

### 4. **Market reach expanded**
- Partnerships with leading distributors (e.g., Koton and Shijie) indicate breakthroughs in channel development, helping expand market share and brand influence.

### 5. **Overall trend is positive**
- Although revenue growth slowed in 2016 compared to 2015, net profit continued to rise at a healthy pace. This signals improved operational efficiency and profitability. It shows the company adapts well to market changes and optimizes effectively.

### Conclusion:
Overall, Hunan Goke Microelectronics Co., Ltd. showed an **optimistic** performance trend from 2014 to 2016. The company achieved consistent growth in both revenue and net profit. It strengthened its core competitiveness and profitability through product mix optimization, innovation, and market expansion. While revenue growth slowed in 2016, stable net profit growth reflects strong operational management and cost control."