An investment thesis is written in one meeting and checked rarely after that. Markets move, companies pivot, and the holding that fit the thesis in January can drift out of it by March. The portfolio review is where the gap should surface, but the review takes days of manual work across holdings, and the thesis lives in a deck that nobody updates, so the drift is found late, when it has already cost returns.
AI thesis assessment keeps the check continuous. The platform reads the market data, the company performance signals, and the unstructured news flow together, scores each holding against the thesis the fund set, and flags drift the day it starts to form.
Shakudo deploys a thesis assessment platform for investment firms. Trino consolidates the financial data across the sources the firm already uses, and dbt transforms it into a clean, auditable analysis layer. PyTorch runs the machine learning models that track market trends and company performance against the thesis. LangChain reads the unstructured data, the earnings calls and news flow, and folds that context into the score. Metabase gives the team a real-time view of how each holding lines up with its thesis, and MLflow keeps the predictive models current as markets shift. The result is a portfolio that is checked against its theses continuously, with drift flagged the day it starts to form, long before the quarterly review. The investment team works from a live alignment picture that updates as the market moves, and the drift shows up the day it forms.
The platform runs in the firm's own environment. Portfolio positions, theses, and trading intent are confidential, and they stay on the firm's infrastructure for the entire workflow. The data pipeline pulls from Trino and shapes it with dbt, the PyTorch models score each holding against its thesis, and LangChain adds the unstructured context from news and filings. The output is a live drift signal the investment team can act on, with Metabase showing the alignment picture at a glance. The signal is a score per holding, so the portfolio manager sees it inside the same review that handles the position itself, and the thesis gets re-checked as the market moves. The quarterly snapshot stops being the first look the team has at the drift. The firm keeps the robustness of big vendor tooling with the flexibility of a custom system, without the build time.
The stack pairs a warehouse-grade data layer with the machine learning tools a quant team already works in. Trino is the data warehousing layer that consolidates the firm's financial data. dbt transforms and models that data into the analysis layer the rest of the stack reads. PyTorch runs the machine learning models for market trends and company performance. LangChain processes the unstructured data, from news to filings, and turns it into signals the model can use. Metabase visualizes the portfolio's alignment with its theses in real time. MLflow tracks and keeps the predictive models accurate and up to date as conditions change.
Investment teams at funds, asset managers, and family offices that hold a thesis per holding and want the fit checked continuously. The setup suits quantitative and fundamental teams in financial services, where portfolio data is confidential and the thesis review today runs on spreadsheets and quarterly cycles.
Each holding is scored continuously against the specific thesis the fund set for it. The PyTorch models track market and company signals, LangChain adds the unstructured context, and the score moves as the evidence moves. A shift that used to be noticed at the next review is flagged the day it starts, with the score and the evidence shown together in the alignment view.
Yes. LangChain reads the unstructured sources, the news flow and the filings, and turns them into signals the models use alongside the structured data in Trino. The thesis score reflects both halves of the market picture, the numbers and the narrative.
Building a thesis assessment system of this kind typically takes months, if not years, of development and integration. Shakudo deploys the full stack within days, so the firm moves from the first integration to live thesis monitoring on a compressed schedule.
For investment teams that check their theses by hand and on a calendar, AI assessment makes the check continuous and the drift signal immediate. Book a demo and see thesis fit scoring run on a live portfolio.