6Tracks
15Problem Statements
1Agriculture 2 Problem Statements
Build technology that can improve the way we grow, distribute, sell, and consume food. Explore challenges faced by farmers, agricultural workers, suppliers, buyers, and rural communities. Think about crop planning, weather and soil insights, irrigation, pest management, farm-to-market connections, agri-finance, supply chains, storage, price discovery, sustainability, and access to agricultural knowledge.
Problem Statements
ps11
PS-01: Soil Health Action Agent
PROBLEM
Small and marginal farmers receive Soil Health Cards containing information about soil nutrients, pH, and fertilizer requirements, but often struggle to convert this technical information into practical decisions for a specific crop and field size.
CHALLENGE
Build an AI-assisted tool that takes a Soil Health Card and a defined crop/field profile, interprets the available soil information, and generates an actionable farming plan.
Example: Soil Health Card → crop + field size → nutrient analysis → fertilizer/input recommendations → application plan.
MVP / EXPECTED OUTCOME
- Soil Health Card upload or structured input
- Crop and field-size profile
- Soil parameter extraction
- Rule-based/AI-assisted interpretation
- Fertilizer/input recommendations
- Local-language action summary
BONUS
- OCR for handwritten/printed cards
- Hindi/Indic languages
- Local units such as bigha/acre
- Cost estimation
- WhatsApp/shareable action card
ps12
PS-02: Mandi Opportunity Agent
PROBLEM
Farmers often have to decide where to sell their produce using fragmented mandi price information. A higher market price may not result in higher earnings once transportation and distance are considered.
CHALLENGE
Build an AI-assisted decision-support tool that takes a farmer's crop, quantity, and location, analyzes market-price information from multiple sources, and identifies potential selling opportunities.
Example: Crop + quantity + location → nearby mandi prices → transport estimation → net realization → opportunity shortlist.
MVP / EXPECTED OUTCOME
- Farmer/crop profile
- At least one market-price data source
- Nearby mandi comparison
- Transport-cost calculation
- Net realization comparison
- Clear opportunity recommendation
BONUS
- Historical price trends
- Multiple data sources
- Map visualization
- FPO/bulk-selling mode
- Hindi/Indic language support
2HealthTech 2 Problem Statements
Healthcare is more than yoga, exercise, and fitness tracking. Think beyond fitness. Build solutions that improve how people prevent, access, understand, and manage healthcare. Explore ideas around preventive care, patient and caregiver support, diagnostics, medication and appointment management, mental well-being, elder care, accessibility, health records, hospital workflows, healthcare operations, and community health. Find an overlooked healthcare problem and use technology to solve it.
Problem Statements
ps21
PS-01: Prescription Understanding Agent
PROBLEM
Patients often leave OPD visits with handwritten or abbreviated prescriptions and may struggle to understand medicine names, dosage, frequency, and timing.
CHALLENGE
Build an assistive AI system that converts an existing prescription into a clear, patient-friendly medication schedule after the extracted information has been verified by the user.
Example: Prescription image → medicine extraction → user verification → dosage/timing schedule → patient/caregiver plan.
MVP / EXPECTED OUTCOME
- Prescription upload
- Medicine and instruction extraction
- Human verification step
- Daily medication schedule
- Patient-friendly explanation
- Caregiver/shareable view
BONUS
- Hindi/Indic languages
- Voice instructions
- Pictograms
- Missed-dose tracking
- Reminder workflow
CONSTRAINT
The system must not diagnose diseases or recommend new medicines.
ps22
PS-02: PM-JAY Hospital Readiness Agent
PROBLEM
PM-JAY beneficiaries may know that they are entitled to healthcare support but still struggle to understand which hospital to visit, what documents to carry, and what to expect during the hospital process.
CHALLENGE
Build an AI-assisted hospital-readiness tool that takes a patient's location and care requirement and generates a practical preparation plan using a defined PM-JAY information source and hospital dataset.
Example: Location + care requirement → suitable hospitals → required documents → visit steps → preparation checklist.
MVP / EXPECTED OUTCOME
- Patient/location profile
- Hospital information source
- Specialty-based filtering
- Document checklist
- Hospital visit workflow
- Plain-language PM-JAY guidance
BONUS
- Hindi/Indic language
- Voice assistant
- ASHA/caregiver mode
- Map-based navigation
- Bounded RAG over official PM-JAY documents
CONSTRAINT
No claims processing, medical diagnosis, or unsupported eligibility/coverage decisions.
3EduTech 2 Problem Statements
Reimagine how people learn, teach, and access education. Build solutions for personalized learning, skill development, student support, assessments, accessibility, educational content, and improving learning outcomes. You can also explore challenges in higher education, such as academic and career guidance, resource and course planning, exam preparation (GMAT, GRE, and other competitive exams), scholarships and elite opportunities, college applications, and helping students discover pathways to their next opportunity.
Problem Statements
ps31
PS-01: Teacher Lesson Planning Agent
PROBLEM
Teachers spend significant time converting textbook chapters into lesson plans, worksheets, and assessment activities while also managing classroom responsibilities.
CHALLENGE
Build an AI agent that takes a defined textbook chapter or learning objective and generates a structured lesson plan and supporting classroom material.
Example: Class + subject + chapter → learning objectives → lesson plan → worksheet → formative assessment.
MVP / EXPECTED OUTCOME
- Class/subject/chapter input
- Defined educational content source
- AI-generated lesson plan
- Worksheet generation
- Formative assessment questions
- Editable/exportable output
BONUS
- Hindi/English output
- NCERT/DIKSHA RAG
- Remedial activity generation
- Parent homework summary
- Voice-based teacher interaction
ps32
PS-02: FLN Learning Gap Agent
PROBLEM
Teachers need to identify specific foundational literacy and numeracy gaps among students, but conducting detailed assessments and manually analyzing the results can be time-consuming.
CHALLENGE
Build a classroom assessment tool that takes student responses from a short FLN assessment, maps them to defined skills, identifies learning gaps, and generates targeted remedial activities.
Example: Student responses → skill mapping → class-level gap analysis → student/group insights → remedial activities.
MVP / EXPECTED OUTCOME
- Class/student profile
- Short predefined FLN assessment
- Response recording
- Skill-level analysis
- Visual gap report
- Remedial activity suggestions
BONUS
- Hindi/Indic language
- Printable assessment sheets
- Parent report
- Progress comparison
- Voice-based assessment
4Startup Support 2 Problem Statements
Help entrepreneurs turn ideas into sustainable businesses and solve the challenges that come with building a startup. Explore solutions for idea validation, customer discovery, market research, fundraising, financial planning, sales, marketing, hiring, team collaboration, productivity, operations, compliance, and business intelligence. Think about the problems founders face before, during, and after launching a product.
Problem Statements
ps41
PS-01: Startup Scheme & Funding Agent
PROBLEM
Early-stage founders often struggle to determine which government startup schemes or seed-funding programs they may qualify for because eligibility requirements are distributed across multiple documents and conditions.
CHALLENGE
Build an AI agent that takes a defined startup profile and uses official scheme documents to identify relevant opportunities, evaluate eligibility conditions, and generate an actionable application plan.
Example: Startup profile → relevant schemes → eligibility analysis → missing requirements → document checklist → action plan.
MVP / EXPECTED OUTCOME
- Startup profile
- At least one official scheme data source
- Eligibility reasoning
- Relevant opportunity identification
- Missing-requirement detection
- Application action plan
BONUS
- Multiple scheme sources
- Document analysis
- Clause-level citations
- Indic language
- Incubator matching
- Application draft generation
ps42
PS-02: Startup Compliance Agent
PROBLEM
Early-stage startups often manage compliance tasks manually and may struggle to understand which recurring responsibilities apply to their company as it grows.
CHALLENGE
Build an AI-assisted compliance agent that takes a defined startup profile and generates a relevant compliance roadmap, including tasks, timelines, responsibilities, and official references.
Example: Company profile → applicable requirements → compliance timeline → task assignment → completion tracking.
MVP / EXPECTED OUTCOME
- Startup/company profile
- Defined compliance information sources
- Applicable-task identification
- 12-month compliance calendar
- Task ownership
- Official source references
BONUS
- Document analysis
- Calendar/ICS export
- Reminder workflow
- Risk flags
- Human approval workflow
- Indic language
CONSTRAINT
The system should act as an organizational assistant, not provide definitive legal advice or automatically file government submissions.
5Social Impact & Prevention 2 Problem Statements
Use technology to address meaningful social challenges and help prevent problems before they escalate. Explore areas such as public safety, awareness, digital well-being, community support, inclusion, youth development, emergency preparedness, accessibility, misinformation, and resource discovery. Think about problems that affect individuals, communities, and society and build solutions that can create measurable positive impact.
Problem Statements
ps51
PS-01: School Zone Safety Agent
PROBLEM
Schools and communities often identify road-safety problems around school gates, such as unsafe crossings, speeding, poor lighting, and parking, but lack a structured way to assess these risks and convert them into actionable prevention measures.
CHALLENGE
Build an AI-assisted safety assessment tool that takes a school location and a structured safety audit, analyzes the identified risks, and generates a prevention plan that can be shared with school management and parents.
Example: School location → safety audit → risk identification → risk summary → prevention actions → community report.
MVP / EXPECTED OUTCOME
- School/location profile
- Road-safety audit checklist
- At least one public/context data source
- Risk categorization
- Prevention recommendations
- Shareable safety report
BONUS
- Map visualization
- Image-based hazard detection
- Before/after audit
- Hindi/Indic language
- Parent/community action workflow
ps52
PS-2: Heatwave Preparedness Agent
PROBLEM
Heatwaves affect households differently depending on factors such as elderly members, children, outdoor workers, housing conditions, and access to cooling. Generic weather alerts do not provide personalized preparedness guidance.
CHALLENGE
Build an AI-assisted preparedness agent that takes a household profile and relevant public heatwave guidance, identifies the household's key risks, and generates a prioritized preparedness plan.
Example: Household profile → vulnerability analysis → heatwave guidance → prioritized actions → preparedness checklist.
MVP / EXPECTED OUTCOME
- Household profile
- Vulnerability factors
- Public heatwave guidance source
- Risk prioritization
- Personalized preparedness checklist
- Shareable guidance
BONUS
- Hindi/Indic language
- Voice interface
- Weather/alert data
- NGO/ASHA worker mode
- Household comparison
- Printable outreach cards
CONSTRAINT
Focus on prevention and preparedness. Do not build a medical diagnosis or treatment system.
6Open Track 5 Problem Statements
Have a problem you care about that doesn't fit into the other tracks? This is your space. Build anything—from an unconventional product idea to a solution for an overlooked everyday problem. Explore technology, creativity, and new possibilities without being restricted to a specific domain. Your problem. Your idea. Your solution.
Problem Statements
ps61
PS-01: Safe Natural-Language Search for Commudle
PROBLEM
People think in sentences, not filters. Commudle users searching for "Android developers near me" or "speakers on Flutter who have spoken in Lucknow" have to translate that into filters and keywords themselves, across communities, events, speakers, hackathons, builds, labs and jobs.
CHALLENGE
Build a pluggable semantic search system that converts a natural-language request into the right query, executes it safely against the database, and returns ranked results. It must not leak private information, and must resist prompt injection from users and from content stored in the database itself.
Example: "find Android developers near me" → intent and filters extracted → validated query against public fields → permission-checked results → ranked profiles.
MVP / EXPECTED OUTCOME
- Search API taking a query plus requester context (logged out, member, organiser, city)
- Natural language to structured query: technology, location, dates, role, content type
- Permission-aware execution against a mock schema with synthetic data
- Injection and leakage defences enforced in code, not in the prompt
- Demo interface showing results and blocked attempts
- Test set: 20 normal queries with expected results, 15 adversarial queries blocked
BONUS
Hybrid search with embeddings plus structured filters, Hindi and Hinglish queries, clarifying questions for ambiguous requests, ranking by recent activity, logging of blocked attempts.
CONSTRAINT
No real personal data. Never expose emails, phone numbers, RSVPs, attendance, form responses, private channels, drafts or organiser-only analytics. The model must never have raw database access.
ps62
PS-02: Merchant Growth AI (Paytm)
PROBLEM
Millions of small merchants run on Paytm without an analyst, a marketer or a finance team. The data to grow their business exists in their transactions, but the insight does not reach them in a form they can act on.
CHALLENGE
Build an AI business partner for a Paytm merchant: something that understands the merchant's transaction history and tells them what to do next, in language a shop owner actually uses.
MVP / EXPECTED OUTCOME
- Merchant-facing assistant working over synthetic transaction data
- Plain-language insights: sales patterns, slow periods, repeat customers, cash flow
- Specific recommended actions, not just charts
- Conversational interface a non-technical merchant can use
BONUS
Hindi and regional language support, voice-first interaction, WhatsApp delivery, proactive alerts, settlement and working-capital insights.
CONSTRAINT
No real merchant data. Must be usable by someone with no analytics background. Must not give regulated financial advice.
ps63
PS-03: AI-Powered Financial Journeys (Paytm)
PROBLEM
Insurance, lending and fintech journeys are full of jargon, forms and drop-off points. Users abandon applications because they do not understand what is being asked or what they are agreeing to.
CHALLENGE
Make one financial journey simpler, faster and more human using AI. Pick a single flow, for example a loan application, an insurance claim, or policy comparison, and rebuild it end to end.
MVP / EXPECTED OUTCOME
- One complete journey working end to end on synthetic data
- Plain-language explanation of terms, eligibility and next steps
- Document understanding where the flow needs it
- Measurably fewer steps or less friction than the current experience
- Clear handling of the moments users usually drop off
BONUS
Indic languages, voice, accessibility for low-literacy users, explainability for automated decisions.
CONSTRAINT
No real financial or personal data. Must not make credit, underwriting or claim decisions autonomously, and must not give regulated financial advice.
ps64
PS-04: Autonomous AI Teammates (Paytm)
PROBLEM
Most AI assistants respond. They do not finish the job. Real work needs something that plans, takes multiple steps, uses tools and closes the loop.
CHALLENGE
Build an AI teammate that completes a real workflow end to end, rather than answering questions about it. Pick one job a person does repeatedly and hand it over entirely.
MVP / EXPECTED OUTCOME
- Agent that plans and executes a multi-step task using tools or APIs
- Visible reasoning and progress, so a human can follow what it did
- Error handling and recovery when a step fails
- Human checkpoints before consequential actions
- A demo showing the task completed start to finish, not a scripted path
BONUS
Multi-agent coordination, memory across sessions, integration with real tools, cost and latency tracking, an evaluation harness for reliability.
CONSTRAINT
No autonomous actions with financial consequences without human approval. No real user data.
ps65
PS-06: AI Signal Intelligence Challenge (MacroVision AI)
PROBLEM
Business and financial decision-makers work across fragmented sources: statements, transactions, invoices, operational records, customer activity, supplier information, reports and contracts. The important signals sit between these sources and only become visible when several pieces are read together, so a human analyst usually spots them late.
CHALLENGE
Build an AI/ML system that discovers meaningful signals, anomalies, trends, relationships and emerging risks or opportunities across heterogeneous business or financial data, and clearly explains the evidence behind each finding.
Example: revenue up 20%, receivables up 55%, DSO up materially, payment terms changed → SIGNAL: working-capital pressure may be increasing → investigate customer concentration, collections and recent credit-term changes.
MVP / EXPECTED OUTCOME
- Ingest at least one structured source (CSV, Excel, JSON, database export, API) and at least one unstructured source (PDF, report, business update)
- Detect anomalies, unusual changes, trends or notable events
- Correlate evidence across multiple records or sources, not a single observation
- Surface risks or opportunities worth human investigation
- Explain each signal with its supporting evidence
- Recommend what a decision-maker should investigate next
- One complete workflow: source data → detected signal → evidence → recommended next step
- Interface or dashboard with an evidence view, plus a technical note on data sources, architecture, methods and limitations
DATA
No dataset is provided. Sourcing it is part of the challenge. Use public or open datasets, public company filings and reports, free-tier APIs within their terms, or synthetic data you generate. Do not rely on confidential, proprietary, illegally obtained or paid restricted data.
BONUS
Choose your own scenario: financial, company, supply-chain, customer, due-diligence or operational intelligence. Techniques worth reaching for include LLM reasoning and extraction, embeddings, semantic search or RAG, anomaly detection, clustering, entity and relationship extraction, temporal reasoning, agentic tool use and predictive modelling.
CONSTRAINT
A dashboard or purely rule-based analytics application is not enough; there must be a meaningful AI/ML component. Do not build an autonomous investment, lending or financial decision engine. The system provides intelligence, evidence and investigation leads for a human, and does not replace human judgement. Software only, no hardware.
