GigaOm, an analyst research firm, produces around 150 in-depth reports annually (50-60 pages each) across various technology areas, including security, data analytics, AI, networking, storage, and cloud infrastructure. As a research and advisory firm, GigaOm's primary role is to clarify the market landscape and assist clients in narrowing down options, rather than providing direct vendor recommendations.


This commitment to impartiality led GigaOm to collaborate with Pythian, who facilitated an AI consulting engagement utilizing the features within Google's Vertex AI and Gemini Model which enhanced the accessibility of GigaOm’s extensive reports. Pythian built a retrieval augmented generation (RAG) system that could synthesize both unstructured text and structured tabular data to provide factually grounded and numerically accurate responses while adhering to GigaOm's requirement of not making specific recommendations.

GigaOm needed to help customers make decisions faster—using AI to summarize their dense analyst reports

The density of GigaOm’s analyst reports (approx. 50-60 pages) delayed customers in their decision making process. To help expedite customer decision making and find the answers to their questions faster, GigaOm believed AI could help them achieve this. However, they quickly discovered that using AI and large language models (LLM) compromised their business. LLMs were providing vendor recommendations.

GigaOm's reports contained complex textual and graphical data, along with numerical information in tables, making it challenging—even with the expert integration of custom prompts—for standard language models to address the issue in a way that would be suitable for GigaOm’s needs.

These reports presented a significant challenge due to their complex nature, incorporating substantial information in both unstructured textual format and structured graphical representations, exemplified by their radar plots. These plots were difficult for even multimodal language models to accurately interpret. Standard language models like Gemini could only effectively process this information in simple, isolated instances. This complexity hindered the ability to extract precise answers to questions posed against the reports.

Pythian built a custom RAG system on Google AI models to summarize responses with analyst research and reports

To tackle the challenge of leveraging AI for this particular use case, GigaOm strategically partnered with Pythian, a Google Cloud Premier Partner. The two organizations had already established a long-standing relationship of trust, founded on similar and synergistic philosophies around facilitating innovative solutions for the customers they serve. For this engagement, GigaOm was keen to tap into Pythian’s expertise in deploying and configuring the AI solutions found within Google Gemini.

To overcome the limitations presented by the complexity of GigaOm’s structured and unstructured data, Pythian developed a custom implementation of a traditional RAG system—building a RAG architecture tailored to GigaOm’s specific needs and use case, rather than relying on a generic pre-built solution. The customization allowed for greater control, enhanced relevance and improved accuracy, especially in dealing with GigaOm’s domain-specific knowledge bases and reports. Pythian used chunking strategies to identify and retrieve relevant information chunks from the reports' textual content in response to a user's query. These targeted chunks were then fed into the language model, enabling it to synthesize concise and accurate responses grounded in the provided context. This approach ensured the system could navigate the complex textual information effectively.

Furthermore, recognizing the limitations of standard RAG systems in handling structured tabular data, Pythian developed an additional custom component. This component was designed to extract pertinent information from the tables relevant to the user's question and integrate it as supplementary context. By synthesizing both the unstructured text retrieved by the RAG system and the structured data extracted from tables, the solution provided a more comprehensive understanding for the generative model, leading to more accurate and factually sound responses than a conventional RAG implementation could achieve.

This innovative partnership between GigaOm and Pythian yielded a successful outcome. The team at GigaOm was encouraged by the effectiveness of the AI tool that had been implemented and by how well it aligned with their analytical principles. They were impressed with Pythian’s ability to tailor the solution so that the AI's responses mirrored the nuanced and objective approach of their own analysts, a testament to Pythian's ability to configure the model appropriately. This successful implementation enabled GigaOm to enhance the accessibility and value of their in-depth research for clients seeking focused insights, all while crucially preserving their impartiality and independence as a leading analyst firm.

• Vertex AI - Gemini models
• Vertex AI - Embedding models
• Vertex AI Notebooks
• Google Cloud Storage
• Langchain
• Langfuse
• Streamlit