TailState · Turning uncertainty into opportunity

Intelligence beyond the average.

AI-native research and products development for complex, uncertain and non-linear systems, turning tail events into decision advantage.

See what we are focused on Our reasoning
Why “TailState”

Average is crowded. The tails are where things get interesting.

Our reasoning

The most important intelligence lives beyond the average.

In anomalies, edge cases, transitions and rare but meaningful states. That is where TailState works.

01

Beyond the obvious

The obvious is already visible, known, or automated. We look beyond it, for asymmetric insight that hasn’t been arbitraged away.

02

Built for the tails

Most systems fail at the tails. We build for them: models designed for the rare states where fortunes and failures are decided.

03

From anomaly to insight

We turn uncertainty, anomalies and system shifts into usable intelligence: evidence where there used to be intuition.

How we work

Two pillars. One direction.

Our R&D pillar explores what is possible. Our product pillar builds what is useful.

Pillar I · R&D Lab

Researching the edge

Explores what is possible.

Our talented AI researchers test frontier ideas before anyone asks them to be profitable. This is the source of everything we ship.

  • New model architectures, datasets and frameworks
  • Frontier work on non-linear and uncertain systems
  • The strongest ideas graduate into products
Pillar II · Product Studio

Building the future

Builds what is useful.

Where bright engineers turn the strongest research into commercial systems people are happy to use: fixing real problems, not chasing novelty.

  • Software, pipelines and decision-support systems
  • Commercial first: real problems, real workflows
  • Enterprise tools, APIs and analytics for B2B customers
Current focus

Two projects in progress.

Both ideas born in the R&D lab. Both are meant for the states where standard systems fail.

Project 01 · Time Series intelligence based on textual data

TailCast: Forecasting the tail events with a time series-textual data foundation model.

We are training, from scratch, a multimodal foundation model that learns the joint behaviour of language, events and time series, working on diffusion and transformer architectures over precisely tagged data.

Ask it a question about the future in plain language. It returns the full conditional distribution of the time series: not a point estimate, not a guess. Potential usecases:

  • →“What if?” scenario analysisPose an event or hypothesis in plain language and read the conditional distribution it implies: expected shift, spread and tail probabilities.
  • →Narrative riskQuantify how the narratives running through news, reports and commentary reshape the distribution of a series, before the numbers move.
Open TailCast

TailCast is live for TailState clients. Sign in with your account, or email info@tailstate.ai to request access and we will share the subscription details.

Scenario engine · conditional distribution of a time series Interactive
Expected shift
…
P(move < −5%)
…
Tail risk vs baseline
…

Illustrative output only. TailState provides research and analytics, not advice or recommendations.

Project 02 · Next-generation agents

Human on the chips.

The best agents out there, are humans (biological agents). Human on the Chips replicates that: a dynamic system of algorithms wrapped around a generative-AI subconscious; an agent that changes how it works based on its personal experiences.

Every AI agent out there today is a static system of algorithms wrapped around an LLM engine, what we call the subconscious. Capable, but fixed: it performs the same on day one thousand as on day one.

  • →Starts like a juniorIt begins by making mistakes: working, taking feedback, correcting course like a new employee.
  • →Grows into a seniorThrough time and feedback it compounds experience into judgment, not just context.
  • →The bottleneck: memoryThat learning curve demands a genuinely new memory-management system. That is the core of our current R&D.
Agent architecture · static vs dynamic
Typical agent
LLM engine
static algorithms · fixed workflow
Human on the Chips
Generative subconscious
dynamic algorithms · adaptive memory

In research & development. Capability shown is the design goal, not a shipped benchmark.

Who we work with

Built for those who own risk.

We work B2B, wherever decisions depend on tail events: anywhere a time series meets the real world.

Healthcare & biotech
Financial institutions & asset managers
Energy & infrastructure
Insurers & corporate risk teams
Technology & data companies
Research & consulting organisations

Every one of these teams already owns risk they cannot fully see around corners on.

About us

At TailState, we build intelligence for the tails.

TailState was founded in London in May 2026, built on a multidisciplinary background across engineering, biotechnology, mathematics and quantitative finance, with hands-on experience in fintech and applied AI.

Alireza Kargarzadeh

Alireza Kargarzadeh

Founder & CEO

MSc Graduate, Imperial College London

Alireza Kargarzadeh brings a multidisciplinary background across engineering, biotechnology, mathematics and quantitative finance. He has hands-on experience in fintech, combining financial markets knowledge with applied AI, and product thinking. His work focuses on turning complex, cross-domain problems into practical research, analytics and technology solutions.

TailState · Turning uncertainty into opportunity

Where things get interesting.

Schedule a time with TailState for an introductory meeting or follow-up discussion, to explore how we can help and identify the most appropriate solutions.

Schedule a meeting info@tailstate.ai