Skip to content
KRIMKAR

Example projects

Build things worth showing.

Labs and the capstone are built around real problems. Projects from the first cohort will be published here as they ship.

Example projects

Build things worth showing.

Every learner leaves with portfolio work. These example briefs show the kind of problems learners take on in labs and the capstone.

AI Agents

Developer track

A support agent that knows when to hand over

Problem
Routine support questions pile up, but a fully automatic bot creates risk on the hard ones.
Approach
An agent with tool access for lookups, clear policy rules, and an escalation path to a human reviewer.
Outcome
A deployed, reviewed agent with a public demo link and a written policy review.
Technology
PythonLLM APITool callingPolicy checks

RAG Systems

Developer track

Answers grounded in your own documents

Problem
Teams lose time searching scattered manuals, circulars and policies.
Approach
Index the documents, retrieve the relevant passages and generate answers that cite their sources.
Outcome
A searchable assistant with citations and an evaluation set of real questions.
Technology
PythonEmbeddingsVector indexLLM API

Data Products

Business track

From spreadsheet sprawl to one decision dashboard

Problem
A business tracks sales and operations across disconnected spreadsheets.
Approach
Clean and join the data with AI assistance, define the few metrics that matter, and build a dashboard.
Outcome
A live analytics dashboard and a one-page decision memo for the owner.
Technology
SpreadsheetsAI-assisted analysisDashboard tools

AI Automation

Business track

An enquiry workflow that runs itself

Problem
Enquiries arrive by form, email and phone and are followed up inconsistently.
Approach
A no-code workflow that captures, classifies and routes enquiries, with AI drafting first replies.
Outcome
A working automation with before-and-after response times documented.
Technology
No-code automationLLM classificationCRM

Voice AI

Developer track

A bilingual voice assistant for first-line queries

Problem
Many callers prefer to speak, and in Hindi as often as English.
Approach
Speech-to-text, an LLM with a narrow, well-tested scope, and speech output, with safe fallbacks.
Outcome
A prototype voice flow tested against a scripted set of real caller questions.
Technology
Speech-to-textLLM APIText-to-speech

AI Evaluation

Developer track

Measuring whether an AI assistant is actually right

Problem
An assistant looks impressive in demos but nobody knows how often it is wrong.
Approach
Build a labelled test set, define pass criteria, and run repeatable evaluations after every change.
Outcome
An evaluation report with accuracy, failure categories and recommended fixes.
Technology
PythonTest datasetsScoring rubrics

Responsible AI

Both tracks

A risk review before an AI tool goes live

Problem
An organisation wants to adopt an AI tool but has no process for assessing its risks.
Approach
Apply a risk assessment covering bias, privacy under the DPDP Act, safety and human oversight.
Outcome
A documented risk assessment and a go-live checklist the organisation can reuse.
Technology
Risk frameworksPolicy templatesRed-teaming

AI Product Prototypes

Business track

A costed AI product plan for a real business

Problem
A local business wants to use AI but cannot judge which idea is worth funding.
Approach
Map the customer problem, prototype the experience, and model costs and return on investment.
Outcome
A clickable prototype and a costed product plan with a clear go / no-go recommendation.
Technology
Product discoveryNo-code prototypeROI model

Your AI career doesn’t start with a job. It starts with capability.

Build the foundations. Learn the systems. Work on real problems.

Apply to Bihar AI LaunchpadBook Free Counselling

Questions? Call +91 84476 13585 or write to admin@krimkar.com.

Free CounsellingApply Now

Apply for Bihar AI Launchpad