Hologres:Build a multimodal AI system for financial data analytics and retrieval

Vector search

qa_vector_search_retrieval embeds the question, retrieves the top 20 semantically similar chunks, reranks them with ai_rank, and passes the top 5 to the LLM.

When to use: The query is conceptual or the exact wording differs from the source document. Vector search excels at intent matching.

-- Vector-only retrieval + AI reranking
SELECT qa_vector_search_retrieval(
  question => 'By what percentage did the operating income and net profit of Goke Microelectronics increase year-over-year in 2014, 2015, and 2016 during the reporting period?',
  corpus_table => 'dt_bs_challenge_financial',
  prompt => 'Please analyze whether the following performance trend is pessimistic or optimistic and provide reasons: ${question}\n\n Reference information:\n\n ${context}'
);

Sample output:

"Based on the information provided, the analysis of the performance trend of Goke Microelectronics leads to the following conclusions:

### I. Performance Trend Analysis: Pessimistic

#### 1. **Sluggish revenue growth**
- Revenue in 2014 increased by **15.13%** year-over-year, but in 2015, it **decreased by 5.21%**. Data for 2016 is not provided, but it is clear that the revenue growth trend saw a significant decline in 2015.
- The compound annual growth rate (CAGR) of revenue from 2012 to 2014 was only **4.47%**, indicating slow business expansion.

#### 2. **Continuous decline in net profit growth**
- Net profit grew by **5.43%** in 2014 but **decreased by 3.29%** in 2015.
- After deducting non-recurring gains and losses, the net profit attributable to parent company shareholders decreased by **3.14%** in 2014 and further reduced by **5.60%** in 2015.
- The CAGR of net profit after deducting non-recurring gains and losses from 2012 to 2014 was **-4.38%**—much lower than the revenue growth—indicating that the main business is not profitable and growth relies on non-recurring gains and losses.

...

### III. Conclusion
Overall, the performance trend of Goke Microelectronics is **pessimistic**."

Full-text search

qa_text_search_retrieval retrieves chunks using an inverted index, reranks with ai_rank, and generates an answer.

When to use: The question contains specific financial figures, company names, or terms that appear verbatim in the documents. Full-text search is faster and more precise for exact-term lookups.

-- Full-text search retrieval
SELECT qa_text_search_retrieval(
    question => 'By what percentage did the operating income and net profit of Goke Microelectronics increase year-over-year in 2014, 2015, and 2016 during the reporting period?',
    corpus_table => 'dt_bs_challenge_financial',
    prompt => 'Please analyze whether the following performance trend is pessimistic or optimistic and provide reasons: ${question}\n\n Reference information:\n\n ${context}'
);

Sample output:

"Based on the information provided, the overall performance trend of Goke Microelectronics in 2014, 2015, and 2016 is **pessimistic**.

### 1. **Sluggish revenue growth**
- Revenue growth rate in 2014 was **15.13%**, but turned to **-5.21%** in 2015.
- The CAGR of revenue from 2012 to 2014 was only **4.47%**.

### 2. **Poor growth in net profit**
- Net profit growth rate in 2014 was **5.43%**, decreased to **-3.29%** in 2015.
- Net profit after deducting non-recurring gains and losses: **-3.14%** in 2014, **-5.60%** in 2015.
- CAGR of net profit excluding non-recurring items from 2012 to 2014 was **-4.38%**.

..."

Hybrid search with reranking

qa_hybrid_retrieval runs vector search and full-text search in parallel (top 20 each), unions the results, reranks all candidates with ai_rank, and selects the top chunk to send to the LLM.

When to use: Queries mix semantic intent with specific terminology. For financial document Q&A, this mode typically produces richer, more structured analysis than either single-mode approach.

-- Full-text and vector dual-retrieval + AI reranking
SELECT qa_hybrid_retrieval(
    question => 'By what percentage did the operating revenue and net profit of Hunan Goke Microelectronics Co., Ltd. increase year-over-year in 2014, 2015, and 2016 during the reporting period?',
    corpus_table => 'dt_bs_challenge_financial',
    prompt => 'Please analyze whether the following performance trend is pessimistic or optimistic and provide reasons: ${question}\n\n Reference information:\n\n ${context}'
);

Sample output (the hybrid mode produced a more structured, multi-dimensional analysis):

"Based on the information provided, we can analyze the performance trend of Goke Microelectronics as follows:

### I. **Operating revenue trend analysis**
1. **2012–2014 CAGR**: 4.47%, with operating income of CNY 181.5406 million in 2014—an increase of 15.13% from 2013.
2. In 2015, operating income **decreased by 5.21%** year-over-year.

### II. **Net profit trend analysis**
1. CAGR of net profit after deducting non-recurring gains and losses: **-4.38%** from 2012 to 2014.
2. Non-recurring gains and losses as a proportion of net profit: **17.54%** (2014), **10.25%** (2013), **8.06%** (2012).

...

### **Final conclusion: overall trend is pessimistic**
The continuous decline in net profit excluding non-recurring items, reliance on non-recurring gains and losses, and large short-term performance fluctuations indicate that performance growth lacks sustainability."

Hybrid search with RRF

qa_hybrid_retrieval_rrf uses Reciprocal Rank Fusion (RRF) instead of a model-based reranker to combine vector and full-text results.

When to use: Use RRF when you want deterministic, model-free fusion. Use qa_hybrid_retrieval when you have a fine-tuned reranking model available.

How RRF scoring works: Each chunk receives a score based on its rank position in both the vector and full-text result lists. A chunk ranked 3rd in vector search and 9th in full-text search gets a score of 1/(60+3) + 1/(60+9) ≈ 0.0159 + 0.0145 = 0.0304. The constant 60 is the rrf_k parameter. A higher rrf_k reduces the weight difference between rank positions, making scores more uniform across the list. The default value of 60 is a standard choice that balances rank position sensitivity without over-weighting the top result.

-- Full-text and vector dual-retrieval + RRF reranking
SELECT qa_hybrid_retrieval_rrf(
    question => 'By what percentage did the operating income and net profit of Goke Microelectronics increase year-over-year in 2014, 2015, and 2016 during the reporting period?',
    corpus_table => 'dt_bs_challenge_financial',
    prompt => 'Please analyze whether the following performance trend is pessimistic or optimistic and provide reasons: ${question}\n\n Reference information:\n\n ${context}'
);

Sample output:

"Based on the information provided, the analysis of the performance trend of Goke Microelectronics leads to the following conclusions:

### **Performance trend judgment: Pessimistic**

1. **Net profit growth is lower than revenue growth:** CAGR of operating income from 2012 to 2014 was 4.47%, but CAGR of net profit after deducting non-recurring gains and losses was **-4.38%**—significantly lower.

2. **Significant fluctuations in net profit:** Net profit from January to March 2015 decreased by 48.26% year-over-year, mainly due to lower gross profit margins on recognized projects.

3. **Projected profit decline in H1 2015:** Estimated net profit of CNY 23.40M–28.60M, lower than H1 2014's CNY 29.1266M.

### **Summary:**
Although operating revenue maintained stable growth, net profit growth significantly lagged or turned negative. If the company cannot increase gross profit margins or control costs, future performance may continue under pressure."