Concept
resilience, or risk reduction and adaptation, provides numerous benefits, but these aren't always as clear as the associated cost of investment. various methodologies and resources exist to quantify and measure them in monetary terms for cost-benefit analysis, but this expertise is not always accessible to would-be users. the system allows a user to define a case and, with agentic support, research values, modify them, run analysis through a deterministic engine, then interpret and present results.
Verdict
the system provides credible results much more rapidly than manual assessment, with traceability and quality assurance documentation; however although pushing the boundary on agentic support it re-emphasises that an informed human user is critical to the process. while there is no such thing as a 'correct' cost-benefit analysis, the evidence produced is valuable for assessing the investment potential of actions. the architecture can be applied universally.
in action
click below to launch one of two interactive demo assessments: Caribbean island coastal resilience, or data centre site selection.
want to assess a resilience investment?
get in touch at the email at the bottom.
emdola market engine
emergent market dynamics of llm-calibrated agents
Concept
markets behave erratically and while there are many factors involved, some irrational looking movement is actually due to the aggregation of coherent but varying individual behaviour. the effects of this on prices are complex to understand, but the emdola market engine presents a new way to model it. 'agent-based modelling' has actually been around for decades, and while in its original formulation is completely distinct from the 'ai agents' of today, LLMs offer a dynamic new way to generate multiple coherent profiles across many dimensions simultaneously. the market engine can be calibrated for different shocks (e.g. geopolitical crises, climate disruptions, economic shocks), producing different sets of retail and institutional investor types with empirically grounded characteristics. these are allowed to interact with each other across numerous Monte Carlo iterations to separate noise from potential price signals.
Verdict
the market engine simulates the emergent behavioural effects of different investors in a market shock across numerous heterogeneous iterations, thereby establishing trends versus volatility across runs. this produces an understanding of market behaviour which may actually be more rational than it appears, where potential mispricing signals emerge and an investment thesis that takes advantage of this information can be formulated. there is no claim that this is predictive of market prices, rather it is indicative of directional pressures from the behavioural tendencies of investors responding to a market shock, independent of, but complementary to, macroeconomic or fundamental drivers. is there a tendency for emergent behaviour to push prices one direction or another after a shock? if so that apparent mispricing pressure may present an opportunity to investigate.
in action
best experienced on desktop
want to try your own scenario?
get in touch at the email at the bottom.
deffii content transformer
democratising economic forecasting for individual investors
Concept
over 100 million people now invest through digital retail platforms. much of this is done without the financial and economic insight that institutional investors have access to. some company-level data is available, but the forward-looking macroeconomic and sectoral trends, scenarios, and what-ifs that underpin a robust investment thesis often are not. what if retail investors had accessible, high-quality economic content to inform their decisions and build their wealth. the deffii economic content transformer applies an agentic system to collect, adapt, and deliver engaging economic content to help retail investors make better-informed financial decisions.
Verdict
a multi-agent architecture means that each agent focuses on a discrete part of the process. this produces materially better outputs than would be generated by prompt interactions with an LLM, and faster and at greater scale than manual curation. it also allows for calibration to user financial needs, and a structured way to process content and produce multiple formats for retail investors. the system still relies on LLM calls to produce and validate content, and while the room for error is reduced by system design it cannot be eradicated (it's grading its own homework), but it can be managed for example by flagging potential issues for review. applying different LLM models to different agents can further minimise the effect. the result is a production-ready architecture, adaptable beyond retail investing to other content domains and user profiles.
in action
want to have a play?
get in touch at the email at the bottom.