Talee — Sense the Innovation
Case studies

Selected work, in production.

A decade of projects across AI, data and software engineering. A few we can talk about — many more under NDA.

Adaptive trading intelligence for a London hedge fund — AI Development transformation visualBeforeAfter
AI Development · 2024

Adaptive trading intelligence for a London hedge fund

+18% Sharpe
Challenge

Signal-stack performance had plateaued. The desk needed a way to combine alternative data sources without overfitting to recent regimes.

Solution

We built a modular signal pipeline with regime-aware ensembling, walk-forward validation and a research portal for analysts to test ideas safely.

Outcome

Eighteen percent Sharpe ratio uplift across the live book. Research-to-production cycle compressed from weeks to days.

Request the full study
Real-time defect detection for an EV battery line — Computer Vision transformation visualBeforeAfter
Computer Vision · 2024

Real-time defect detection for an EV battery line

99.2% recall
Challenge

Cell-level defects were escaping a traditional rule-based vision system, costing the manufacturer in warranty claims and recalls.

Solution

A multi-camera vision system on edge GPUs, trained on annotated defect data with active learning to capture the long tail of rare defects.

Outcome

99.2% recall at sub-100ms inference. Escape rate cut by 12×, with confidence-based human review on borderline cases.

Request the full study
Clinical decision graph for a national health provider — Knowledge Systems transformation visualBeforeAfter
Knowledge Systems · 2023

Clinical decision graph for a national health provider

−35% diagnosis time
Challenge

Clinicians were navigating dozens of guideline documents per case. Time-to-diagnosis varied widely between regions.

Solution

A clinical knowledge graph linking symptoms, conditions, guidelines and local protocols, surfaced in the existing EHR through a thin assistant layer.

Outcome

Thirty-five percent reduction in diagnosis time on tracked pathways. Clinician satisfaction scores up across all pilot sites.

Request the full study
Petabyte-scale lakehouse for a telecom operator — Big Data transformation visualBeforeAfter
Big Data · 2023

Petabyte-scale lakehouse for a telecom operator

8× query speed
Challenge

Network and customer data were locked in disconnected warehouses. Analytics teams waited hours for queries that should take minutes.

Solution

A unified lakehouse on Delta with streaming ingestion, semantic modelling in dbt and BI access through a governed serving layer.

Outcome

Eight-times faster median query speed. Three legacy warehouses decommissioned. Annual platform cost down by a third.

Request the full study
Multilingual sentiment platform for global retail — NLP transformation visualBeforeAfter
NLP · 2022

Multilingual sentiment platform for global retail

12 languages
Challenge

Brand and product sentiment was tracked manually per region. The team had no consistent view of how launches performed globally.

Solution

A multilingual sentiment pipeline covering twelve languages, with topic clustering and a reviewer console for ambiguous cases.

Outcome

Real-time global sentiment dashboards for marketing leadership. Issue-detection lead time reduced from days to hours.

Request the full study
Document AI for a public-sector archive — Image Processing transformation visualBeforeAfter
Image Processing · 2022

Document AI for a public-sector archive

12M docs / mo
Challenge

A national archive needed to digitise and classify legacy paper records — including hand-written, multilingual and badly scanned documents.

Solution

A multi-stage document AI pipeline combining OCR, layout analysis, classification and a human-in-the-loop review tool for low-confidence pages.

Outcome

Twelve million documents processed per month. Catalogue access for the public years ahead of the original schedule.

Request the full study
Under NDA

There's more we can't show publicly.

Most of our finance, healthcare and public-sector work sits under NDA. Tell us about your problem and we'll share the relevant references on a call.

Request references
FAQ

Case study questions, answered.

06 · Questions

01

Can we get references from clients in our sector?

+

Yes — once we understand your problem and sign mutual NDAs, we will arrange a reference call with a comparable client where possible. Most of our finance, healthcare and public-sector work is otherwise private.

02

Are the metrics on this page independently verified?

+

All numbers are taken from client-side measurements agreed with us at project close — most have been reviewed by the client's data or analytics team. We do not publish self-attested figures.

03

Why are most projects under NDA?

+

We work in regulated and competitive sectors where the existence of an AI capability is itself sensitive information. NDAs protect both the client and our other clients in the same market.

04

What was the typical engagement length for these projects?

+

Discovery is two weeks. Build engagements ran between 3 and 9 months depending on scope. Most of these projects continue today as managed retainers or embedded squads.

05

Do projects continue after launch?

+

Usually yes. We move the system into a managed retainer with SLAs, on-call coverage and quarterly reviews, or hand it over cleanly to the client's in-house team with full documentation and a transition plan.

06

Can you build something similar for us?

+

Almost always. The underlying patterns — vision QA, document AI, predictive analytics, knowledge graphs — generalise across sectors. The work is in the adaptation: your data, your constraints, your tooling.