Thinkscoop Engineering
We build AI systems that survive contact with real data: retrieval you can measure, evaluation in your pipeline, guardrails on the output path, and a cost per task somebody can point at. In your cloud account, in your repository, with the IP assigned to you from the first commit.
engineering / engagements
The price is on the page because an engineering leader with a real project should not have to sit through a call to find out whether the number starts with a two or a two hundred.
engineering / architecture
Different problems, similar architecture. The interesting part is not the model call in the middle, it is the permission that travels with a document into the index, the state you can replay, the escalation that arrives with its evidence assembled, and the layer along the bottom that tells you when any of it starts to drift.
engineering / capabilities
Every capability page names what we instrument and where the approach breaks down. A buyer who finds the limits here is not surprised by them in month three.
C1
Getting the right passages in front of the model. Most quality problems that look like reasoning problems are retrieval problems.
7 practicesC2
A labelled set, a threshold agreed in writing, and a gate in CI. Built before the feature, not after the first complaint.
8 practicesC3
Explicit state, typed tools, bounded loops and a replayable trace. An agent is a distributed system, so we build it like one.
8 practicesC4
What the system may do alone, what it must hand to a person, and what the person receives when it does.
8 practicesC5
Quality, latency and cost per task on a dashboard your team owns, with an alert when any of the three moves.
8 practicesC6
The application around the model: interface, data model, auth, pipelines, infrastructure as code and a deployment your team can run.
8 practicesengineering / delivered
These are Thinkscoop engagements that shipped. The figures are the ones the clients agreed to publish, and nothing here is rounded up for a website.
EY (Ernst & Young)
72%reduction in report preparation timeA multi-agent AI system operating entirely within EY's Azure environment that ingests audit data, cross-references regulatory frameworks, and generates source-cited draft reports for human partner review.
12 weeksBooking.com
68%queries resolved without human interventionA LangGraph reasoning agent with live access to booking APIs, a RAG knowledge base over 500,000+ policy documents, and direct payment system integration - with confidence-based escalation that assembles full context for human agents before routing.
10 weeksGlobal FinTech (Confidential)
78%reduction in processing timeA multi-step LangGraph reconciliation agent that reasons over discrepancy reports, queries source systems via API, and drafts resolution actions for human review - with a full evaluation harness measuring accuracy and escalation rate weekly.
10 weeksHindustan Unilever Limited
4.2minavg query resolution (was 2 days)A unified AI knowledge assistant with RAG over 200,000+ internal documents, role-based access control tied to SSO, full audit logging, and citation-grounded answers - operating entirely within HUL's Azure environment.
14 weeksSAMCo
3×faster portfolio risk analysisA real-time portfolio intelligence platform with live market data feeds, AI risk modelling, and a natural language query interface - so managers get cited answers to risk exposure questions in seconds, not a day later.
8 weeksPixis
35%reduction in ad spend wastageA multi-touch attribution engine with AI-assisted causal inference, processing 5M+ events daily across Google, Meta, and TikTok - giving marketing teams a defensible, explainable model for budget decisions.
6 weeksengineering / standards
These are not aspirations. They are the conditions under which we take the work, and they are the reason some of it we do not take.
E1
No implementation begins against an unmeasured target. If your domain experts cannot spare the time to label a set, that is a real constraint and we plan around it, but we will not substitute a vibe check and call it quality assurance.
E2
Your cloud tenancy, your repository, your credentials, issued by you and revocable by you. Nothing we build depends on infrastructure we control, so the end of an engagement is a handover rather than a migration.
E3
One interface for every model call, prompts versioned as artefacts rather than scattered string literals, provider selection as configuration, and an evaluation suite that can run against more than one provider. Changing provider should cost a configuration change and an evaluation run.
E4
A lookup table, a query or a rule is not a worse answer for being unfashionable. We use a model for the part of the problem that genuinely needs judgement and ordinary code for the rest, which is usually most of it.
E5
We work in your repository, on branches, through your review process. If your team cannot review what we wrote, we have handed you a dependency rather than a system, whatever the demo looked like.
E6
Every engagement scope names what it excludes. Every capability page names where the approach stops working. A buyer who finds the limits on the website is not surprised by them in month three.
The full method, the stage by stage plan and the list of work we turn down are on how we work.
An engineer reads it, not a sales desk. You get a written view of the problem within one working day, including the version where the answer is that you should not build this.