Simon-Kucher & Partners

Consultant — Pricing & Analytics • October 2024 - March 2025 (6 months)
Consulting project • Geneva, Switzerland

My Internship at Simon-Kucher: Building AI Solutions for Private Banking

I recently completed an internship on the Generative AI team at Simon-Kucher, a consulting firm specializing in private-banking pricing. During my time there, I contributed to several production-grade and prototype initiatives, applying strong software engineering practices, rapid learning in LLM tooling, and consulting judgment in a banking context. Here’s what I worked on:

1. Survey Builder (LangGraph, OpenAI API)

I co-developed a LangGraph application that generates market surveys from concise inputs—such as a title, objectives, and product data—while adhering to consulting best practices. My contributions included:

  • Implementing an external research node to enrich survey plans with relevant data.
  • Adding survey-plan comparison functionality to help teams evaluate options.
  • Standardizing outputs (JSON → Excel) for seamless client review.
  • Elevating engineering quality by integrating pytest into pre-commit, enabling linting and commit hooks to ensure every change met industry standards.
  • Creating a short video demo and documenting design choices (e.g., keeping inputs minimal and letting the system infer needs), which helped stakeholders understand trade-offs and accelerated adoption.

2. Private-Banking Portfolio Analyzer

For our pricing/fee-assessment tool used by relationship managers, I tackled the challenge of LLMs and numerical accuracy. My work included:

  • Separating computation from generation to avoid errors.
  • Using structured outputs for fee items (bps, percentages, amounts) to prevent unit drift.
  • Introducing checks to reduce confusion around “negative leakage” and other numeric pitfalls.
  • Auditing report math and units, enriching the tariff section, and improving JSON→Markdown reporting.
  • The result: clearer alerts for incoherent or critically low fees, and more reliable guidance for bankers.

3. RAG Knowledge Assistant

I enhanced our prototype with:

  • Metadata-rich indexing (e.g., URL, document name, stable IDs) to improve retrieval accuracy.
  • Prompt integration of metadata to ensure faithful retrieval.
  • A “document finder” mode—a file-locating agent—for cases where classic RAG was overkill, streamlining day-to-day knowledge discovery for consultants.

4. RFQ Extractor (LangExtract)

I built a lightweight pipeline to extract key RFQ fields, including:

  • Creating a small sample dataset for evaluation.
  • Iterating on extraction prompts to establish a reliable baseline.
  • Helping the team gauge feasibility for structured capture of deal information.

5. Voice Agent (OpenAI Realtime API + LangGraph)

I prototyped a training agent that simulates an M&A negotiator. My work included:

  • Adapting an agent graph and defining a clear system role for the counterpart.
  • Persisting transcripts to simple JSON for easy review.
  • Implementing an end-of-call critique node that summarizes outcomes and provides performance feedback—useful for training bankers in negotiation dynamics.

Engineering & Collaboration

Across all projects, I maintained professional engineering habits:

  • Version-controlled branches with clean merges.
  • Tests and linting in pre-commit to ensure code quality.
  • Readable prompts/config and pragmatic documentation.
  • Clear communication, asking the right questions, and focusing on consulting-grade outputs (clarity, actionability, client-safe wording).

I also quickly got up to speed on finance topics—fees, tariffs, leakage, ROA—and applied this understanding to make models more practical for real banking workflows.