How our Claude Developers Built a Portfolio Intelligence System for Corbin Capital

TL;DR: Claude developers for financial services isn't just positioning. It's the pattern GoML used to build Corbin Capital's portfolio intelligence copilot: a document-aware AI system that cut research on 43+ major holdings from months to minutes, delivered 80% faster answers to senior portfolio leaders, and shipped in a 6-week production build. The RAG-and-agent architecture ports directly to Claude on Amazon Bedrock.

Corbin Capital runs bespoke portfolios across alternative asset classes. 43+ major holdings, some over $100M, each with its own performance history, structural details, and compliance context buried inside PDFs, spreadsheets, and scanned reports.

Portfolio managers used to spend weeks pulling those threads together for a single client conversation. Not anymore.

The Problem

Corbin's edge is tailoring strategies asset-by-asset. That requires senior portfolio leaders to know each holding cold — performance history, structural details, risk profile, contract clauses.

At scale, that broke down.

• Manual document dives ran into weeks per asset.
• Historical context sat in siloed reports no one could search.
• Cross-asset comparisons required someone to hold every detail in their head at once.
• Client advice slowed as portfolios grew.

Corbin needed a way to make every senior leader as informed as the analyst who last touched a given asset.

The Solution

GoML's Claude developers designed and built a document-aware AI copilot for Corbin's portfolio team. The system ingests structured and unstructured portfolio data — PDFs, spreadsheets, scanned reports, audio transcripts — and lets managers query it in natural language.

The original build used GPT-4 Turbo on Azure, the frontier model that met Corbin's quality bar at engagement time. The architecture is model-agnostic and ports cleanly to Claude on Amazon Bedrock for equivalent workloads today.

What the Copilot Does

• Natural-language queries across the portfolio. "Compare 3-year returns of Holding A vs Holding B" or "List all assets with over 20% volatility last year" returns structured answers in seconds.
• Cross-portfolio comparisons. Surfaces differences in asset structures, risk profiles, and historical performance side by side.
• Mathematical modelling. Sharpe ratios, CAGR, and other functions execute in-chat with outputs embedded into PDF-ready reports.
• Compliance-grade controls. Authentication, role-based access, and full audit trails meet financial services requirements.

The Architecture

Data Layer

Portfolio documents ingested from SharePoint and internal repositories. Preprocessing for PDFs, scanned documents, tables, and multimedia. Indexed and stored in Azure Cosmos DB with secure access.

Intel Layer

Retrieval-augmented generation pipeline over embedded document vectors. Vector search and retrieval wired to the frontier LLM. Role-based access controls and usage tracking baked in.

Consumption Layer

React frontend on a secured Linux VM. Real-time queries, comparisons, and custom report generation. Usage analytics and feedback loop for continuous improvement.

The Impact

Corbin's senior portfolio leaders now get answers in minutes rather than months. Reclaimed time goes back into client advisory, not document search.

• 80% faster portfolio research. Manager time reallocated from document search to client work.
• Zero information silos. Every senior leader queries the same knowledge base.
• Fewer errors. Automated retrieval and calculation removes manual copy-paste from the loop.
• 6-week production build. Requirements to deployed copilot.

"GoML helped us unlock the intelligence within our documents. Now our portfolio teams spend less time searching and more time advising."

— Senior executive, Corbin Capital

What Other Alternative Asset Firms Can Take from This

Pitfalls to Avoid

• Relying only on spreadsheets for high-value portfolio insights.
• Underinvesting in data governance and access control on top of the AI layer.
• Building tools without domain-specific financial NLP capability.

What to Prioritise

• Start with the highest-impact use cases: portfolio comparisons and reporting first.
• Pick a partner who understands both LLM engineering and financial compliance.
• Design for scale and analyst-grade usability from day one.

Build the Same Pattern on Claude

The RAG-and-agent pattern behind Corbin's copilot ports directly to Claude on Amazon Bedrock, with GoML's Claude developers running the build. If your firm is sitting on a document-heavy portfolio and looking to unlock it, we can scope a similar system on Claude in a 4-day POC.