Skip to content

Home · Case studies

AI case studies.

AI assistants, agents and tools we have built and run. Names are left out on purpose. Ask us and we will walk you through any of them.

Featured · Education · Live

AI for a preschool: an assistant for parents and a workspace for teachers

Problem

Parents asked the same questions about timings, programmes and admissions at all hours, in English and Hindi. Teachers spent their evenings writing lesson plans, newsletters and parent messages from scratch.

What we built

A parent assistant on the school website that answers only from the school's own pages, in English or Hindi, and shows the page each answer came from. It never guesses fees or gives medical advice; when unsure it hands over to WhatsApp, and callback requests go straight to the school by email. Teachers sign in to a separate workspace with guided tools for lesson plans, weekly curriculum, activities, observation notes, newsletters and parent messages, with roles for teachers, centre heads and directors and plans shared across branches.

Retrieval from the school websiteOpen-source modelsRole-based sign-inWord and PDF export
24/7Answers for parents, with a human hand-off
EN + HIEnglish and Hindi, chosen by the parent
9Guided tools for teachers
4Staff roles, from teacher to admin

More AI work

Six more systems, in plain words.

Knowledge assistant · Live

Answers from hundreds of published articles

Problem

A technical website had hundreds of articles. Readers could not find the one that answered their question.

What we built

An assistant that reads every post and page, refreshes itself every night, and answers with a link to the source article. It says so when the site does not cover a question instead of making something up.

NightlyAutomatic refresh
SourceLinked with every answer
₹0Model licence fees
Retrieval (RAG)WordPressOpen-source models
Document search · Live

Ask questions of 9,000 pages of product documentation

Problem

Engineers needed answers buried in product guides running to thousands of pages, and searching them by keyword was slow.

What we built

A document assistant that splits each guide into passages, stores them in a vector database and answers with the page number each answer came from. New guides are added without rebuilding the rest.

9,000+Pages indexed
33,000+Searchable passages
Page no.With every answer
Vector databaseLocal embeddingsOpen-source models
Training · Live

A studio that writes technical training material

Problem

Trainers for cloud, VMware, Kubernetes and AI infrastructure courses rebuilt slides, labs and quizzes by hand for every batch.

What we built

A signed-in studio where trainers pick a tool, fill a short form and get course material, diagrams, quizzes or certificates to review and export. Admins manage teams, tracks and attendance, and see who changed what.

18Training tools
5Learning tracks
LoggedAdmin actions
Role-based accessDiagram generationAudit log
Finance · Live

Plain-English stock research for everyday investors

Problem

Charts and ratios are hard to read for people who are not traders, and generic chatbots make up numbers.

What we built

A research app that calculates technical and fundamental figures first, then asks the AI to explain only those figures. It never gives buy or sell calls. Members get screeners, watchlists and research briefs.

2,000+NSE stocks searchable
NoBuy or sell tips
GroundedIn calculated data
Market dataPaid membershipOpen-source models
Content · In use

One topic in, ready posts for every platform out

Problem

Posting regularly about public topics such as exams, schemes and city planning took hours of reading and writing each week.

What we built

Type a topic and the tool gathers recent news and official government releases, then writes an X post and thread, an Instagram caption and Facebook, LinkedIn and WhatsApp versions from the same facts. A person reviews every post before it goes out.

6Formats from one topic
DailyScheduled packs, free to run
HumanApproves before posting
News and government sourcesScheduling
Agent · Prototype

An agent that checks travel expense claims

Problem

Finance teams check every travel bill against a long policy by hand, and decisions vary from person to person.

What we built

An agent where the AI reads the bills and the policy, but a fixed rules engine decides every amount. Each line comes out approved, rejected or held for a person, with the policy clause that applies.

3Outcomes: approve, reject, hold
ClauseCited with every decision
18Automated tests
Agent designRules engineAzure-ready

We only quote numbers we can show you in a walkthrough.

Want something like this?

Talk to us

Have a project or a problem? Talk to our team.