Agentic AI for Finance: First Impressions of Nicole Königstein’s Packt Course

Gifted review ticket from Packt Publishing — honest, unpaid opinion below.

I completed the first two of four 4-hour live sessions in Nicole Königstein's "Agentic AI for Finance" certification course with Packt, and want to share, as an educator and researcher in quantitative finance, an honest first impression, with more sessions still to come.

The instructional design is strong. The course builds a coherent "harness engineering" vocabulary — state, tools, retrieval, verification, and governance around a model call — applied consistently to filings, earnings calls, and risk workflows. Day two's Session 3, on time series and forecasting, was the exceptional highlight for me: tokenization strategies (point-wise, patching, lagged features, quantized bins, and digit strings), five model families — Chronos, PatchTST, TimesFM, Lag-Llama, and AnomalyBERT for self-supervised anomaly detection — and the framing of "Risk Monitoring as an Event-Driven Loop" and "Reasoning Over Series, News, and Documents" connect modern AI tooling to problems a quant desk actually has. The point that a forecasting model is a tool call inside a harness, not the application itself, is well made. Nicole also shared plenty of extra resources on the fly — LangChain, LiteLLM, OpenRouter, Docling, Tavily, the AnomalyBERT paper and repo, among others — a practical layer beyond the slides.

One respectful reservation, given the title "Agentic AI for Finance": much of both decks — tokenization/embedding recap, ViT and diffusion-transformer mechanics, the audio pipeline from waveform to Whisper — is general AI/ML fundamentals, with no finance framing; it would suit any agentic-AI course. Finance appears explicitly, and well, in the applied sections: harness examples, counterparty exposure, earnings-call extraction, risk monitoring. But tying a specific technique to a specific financial problem end to end happens mainly in two large hands-on notebooks (a 10-K financial-document agent; a multimodal assistant on the Fed's Financial Stability Report). Each day also assigns many smaller session notebooks — RAG, LangGraph, harness components, Chronos/PatchTST/TimesFM, multimodal and audio — useful, but like many slides, technique-focused rather than distinctively finance-specific — with one exception, a notebook built around financial sentiment classification. Given the scope, working through it properly is realistically left to participants.

None of this changes my overall view: the material is well structured and current, the live sessions were clear and well-paced, and both reflect Nicole's proficiency. I look forward to the remaining sessions before a final judgment, though what genuinely excites me more is digesting this worthwhile material and putting it into practice in my own domain.

Curious to join the final two sessions? A pass for Days 3 & 4 live, plus recordings of all four, can be purchased here.