Nimble decision-making powered by agentic workflows for leading asset management firm
While Claude’s reasoning capabilities and domain expertise provide the engine, Proxet’s architectural expertise transforms the engine into a high-performance investment vehicle. By building custom agentic workflows, with expert data optimization pipelines and rigorous data validation, Proxet is enabling a leading global investment firm to move human teams from manual data processing to high-impact analysis and strategic execution. Throughout the process, Proxet leverages Anthropic technology to write our code and build Claude-powered applications, and advised working with Claude for its powerful performance and advanced context reasoning.
Agentic AI for Document Analysis
Transforming unwieldy, unstructured narratives into actionable investment signals
Challenge:
- Analysts manually evaluated investment suitability by cross-referencing owner documents with multiple external databases to produce hundred-page documents–consuming precious analysis time in the deal-sourcing pipeline.
- The volume of information and hundreds of data points per document required human teams a week to process a portfolio of sites.
- Huge risk of missing either crucial investment opportunities or high-risk factors buried in legal and financial prose.
- Classical approaches to this visual and unstructured data weren't enough, requiring heavy lifting to collect labelled data. Also, the solution would have had to involve multiple custom models.
Solution (Proxet + Claude):
- Proxet deployed an agentic workflow using Claude Agent SDK to parse these complex PDFs and interpret highly visual content, including maps and topography.
- Validated and optimized output based on historical data to ensure accuracy.
- Harnessed Claude’s multimodal capabilities to automatically locate, crop, and extract specific images buried within hundreds of pages.
Outcome:
- Reduced portfolio analysis time from one week to several hours.
- Shifted analyst time from manual data processing to deeper strategic analysis and decision-making to move quickly on business opportunities.
- Strengthened data fidelity in the resulting output document and included tabular source tracing for analyst verification.
- With Claude, we enabled a single, highly orchestrated multimodal agent for diverse extraction tasks.
Real-Time Custom Market & News Digest via Agentic AI
Synthesizing global industry, legal, political dynamic noise into a daily investment briefing
Challenge:
- Time-consuming human tracking of signals across dozens of global news and regulatory sites was not real-time and was prone to information gaps.
- Access restrictions to licensed, paywalled, or blacklisted sources are required for comprehensive market awareness.
- Delays in updates put investment decisions in jeopardy.
- Hundreds of sites and the volume of data to review caused LLMs to skip critical items during iteration.
Solution (Proxet + Claude):
- With Claude Agent SDK, Proxet developed two specialized sub-agents running in parallel–one for general news and one for financial reports–to specialize searches.
- Combined LLM processing with deterministic algorithms to ensure every list item was covered (no “laziness” or omissions).
- Wrapped external web search providers as custom tools to allow the agent to bypass paywalls and access restricted content.
Outcome:
- Automated a daily digest comparing current news against previous records to highlight only the freshest updates.
- Achieved faster signal detection, picking up market changes as soon as they are published.
- Provided a holistic market understanding by ingesting proprietary analysis into a single, automated digest.
- Automated additional context and confidence for investment decisions in the previous section.
Precise Agentic Standardization of Commercial Contracts and Financial Audit
Ensuring accuracy and ease of analysis for regular reporting to investors from fragmented, multi-format data
Challenge:
- The influx of extensive excel spreadsheets across multiple entities for analysis had no standard terminology, format, or quality control.
- Terminology for critical profitability fields varied wildly between different managers or was missing.
- Repetitive, boring manual entry resulted in frequent copy-paste errors and data inconsistencies.
Solution (Proxet + Claude):
- Implemented an agent capable of symbolic computation and decision-making to infer column meanings by performing mathematical checks.
- Utilized Claude Agent SDK to write custom code on the fly for parsing and file manipulation.
- Standardized diverse reporting formats into a single, predetermined schema provided by the human stakeholder.
- Created automated validation data set and steps, which Claude improved, to test the pipeline and ensure extraction matched historical ground truth (i.e. the summary column at the end of the spreadsheet).
Outcome:
- Processing time for standardization was reduced from hours to minutes.
- Significantly increased accuracy and confidence for sensitive financial reporting.
- Ensured full auditability through agentic traces that show the thinking process and actions undertaken for every record.
- Through Claude, ensured strict security requirements via a zero-data retention policy for financial data handling.