Applied AI research · London
Find out what AI
can do for you.
We test AI ideas, build prototypes and evaluate answers, so you can decide what deserves further investment.
Business software and human review · Concept illustration
- FeasibilityIs the idea worth testing?
- PrototypingDoes the essential approach work?
- EvaluationWhere do the answers fail?
A clear question.
A useful next step.
Engagements
Evidence for your
next decision.
Know whether the idea is practical
Examine the task, available data and simpler alternatives before investing in a prototype.
You receiveA feasibility recommendation and an experiment brief.
Explore feasibility
Test the essential behaviour
Build a limited prototype around a defined question, with a baseline and criteria agreed before testing.
You receiveAn inspectable experiment, findings and remaining uncertainties.
Explore prototyping
Understand the failures
Compare model outputs, check their supporting evidence and examine the mistakes an average score can hide.
You receiveA documented evaluation with examples and a reasoned next step.
Explore evaluationSee the method
Read the answer.
Then check its evidence.
A correct statement can sit beside an invented one. Our worked example shows how to tell them apart.
Inspect the example and worksheet ↗The source says
Changes to the policy require approval from the operations lead before publication.
The answer adds
“The operations lead and finance director approve changes.”
The extra approver is unsupported. A plausible answer still needs a claim-by-claim check.
Invented policy and authored answer. Not a client result or model benchmark.How we work
Start with a decision.
Agree what would inform it.
The question, permitted material, deliverables, fee and review dates are agreed before work begins.
How an engagement runs ↗- 01
Frame the question
Define who needs the answer, the constraints and the evidence that would change the decision.
- 02
Investigate and test
Examine the material or build the agreed experiment. Record the method, failures and limits.
- 03
Make the next step clear
Proceed, narrow the task, test again or stop. Hand over the reasoning and agreed artefacts.
Leadership
A named person.
A defined piece of work.
Manas Labs is led by Shilpa Yadav, Director & Chief Architect, and is based in Stanmore, Greater London.
We work with product teams and domain specialists who need to understand an AI approach before deciding what to build or change. The work is scoped around the question, not a preferred model.
Shilpa Yadav Director & Chief Architect
About Manas Labs ↗Inside the lab
Questions we are
exploring.
Internal concepts and reference work, with their current status stated explicitly.
Private RAG
Examining how selected source passages reach an answer, and where retrieval and source support need separate checks.
Explore the architecture ↗EvidenceDesk
A proposed interface for inspecting claims alongside their sources and identifying missing support.
Read the concept ↗
Sentinel
Investigating evidence and procedure review around laboratory incidents. Not an operational incident-response system.
Explore the question ↗Future research direction
Exploring quantum
computing.
We are shaping questions around scheduling and resource allocation for a possible future research proposal. No quantum experiments, validated results or funding award are being announced.
Read the proposed direction ↗Practical guides
Methods you can
examine.
What should an AI feasibility study establish?
Read the guide ↗What should your first prototype prove?
Read the guide ↗How do you check whether an AI answer is supported?
Read the guide ↗What does Manas Labs do?
We provide applied AI research, feasibility studies, focused proofs of concept and model evaluation. The aim is evidence for a defined technical or investment decision.
Do I need a technical brief?
No. Start with the task, who needs the result and an example of a useful outcome. We can help define the question and the material needed to investigate it.
Will every project lead to a build?
No. A useful investigation can recommend a simpler approach, another test or stopping. The evidence and remaining uncertainty should make that recommendation understandable.
Who handles production implementation?
The findings can go to your own delivery team or a chosen implementation partner. Softomate Solutions is the implementation business for software and business automation; those engagements are separately scoped.
What happens after I enquire?
We reply to clarify the task and available material, then discuss whether a scoped engagement is appropriate. No work begins until the scope, fee and terms are agreed.