Skip to content

DemoScript.md

Latest commit

 

History

History
172 lines (114 loc) · 7.57 KB

DemoScript.md

File metadata and controls

172 lines (114 loc) · 7.57 KB
Raw

Manak AI — SIH Demo Script

3-5 minute demonstration flow for SIH judges

Demo Overview

Duration: 4 minutes Format: Live demo on deployed URL + brief Q&A

Opening (30 seconds)

"Namaste! We are [Team Name]. We built Manak AI — India's first AI-powered BIS Compliance Intelligence Platform.

India has over 19,000 Indian Standards. Determining which ones apply to your product, whether certification is mandatory, and what the actual requirements are — is confusing. Especially for small manufacturers and consumers.

Manak AI solves this. Let me show you."

Demo Query 1: Manufacturer Query (60 seconds)

Input

I manufacture stainless steel water bottles. Which BIS standards and certification requirements apply to my product?

What to Highlight for Judges

  1. Entity Extraction: "Notice how the system identified the product (stainless steel water bottle), material (stainless steel), and user type (manufacturer)"

  2. Citation Quality: "Every claim is backed by a specific clause and page number from an actual Indian Standard — IS 14543:2016, Clause 4.2, Page 12"

  3. Mandatory vs Voluntary: "The system correctly identifies this product is under a Quality Control Order — BIS certification is MANDATORY, not voluntary. It tells you the specific QCO notification number."

  4. Compliance Pathway: "Below the answer, you see a structured compliance pathway — step by step, from identifying standards to applying for ISI Mark certification"

  5. Confidence Level: "The confidence indicator shows HIGH — because we found strong, authoritative evidence from multiple clauses"

Demo Query 2: Consumer Query (60 seconds)

Input

I want to buy a pressure cooker. How do I know if it is BIS certified and safe to use?

What to Highlight

  1. Consumer-Friendly Language: "Notice the response is in simple language — not technical jargon. It explains what ISI Mark is, where to find it on the product, and how to verify it"

  2. Practical Information: "It tells the consumer about IS 2347:2017 (Domestic Pressure Cookers), what safety tests are required (burst pressure, safety valve), and how to check ISI Mark using the BIS Care app"

  3. Citation Panel: "On the right, you can see all source citations — each one verifiable"

Demo Query 3: Hindi Query (60 seconds)

Input

मुझे खिलौने बनाने हैं। कौन से BIS मानक लागू होते हैं? क्या BIS प्रमाणन अनिवार्य है?

(Translation: I want to manufacture toys. Which BIS standards apply? Is BIS certification mandatory?)

What to Highlight

  1. Hindi Response: "The system detected Hindi and responded in Hindi — while preserving standard numbers like IS 9873 in their original form"

  2. Multilingual Retrieval: "Even though our knowledge base is primarily in English, the multilingual embeddings correctly retrieved the relevant toy safety standard"

  3. Accurate Mandatory Status: "It correctly identifies that toys under IS 9873 are under mandatory QCO by DPIIT — with the effective date"

Demo Query 4: Safety/Failure Demo (45 seconds)

Input

What are the BIS standards for artificial intelligence software?

What to Highlight

"This is our safety feature. Watch what happens when we ask about something NOT in our knowledge base."

  1. Honest Refusal: "The system says: 'The available documents do not contain specific Indian Standards for AI software. While BIS may be developing standards in this area, I cannot provide requirements without authoritative evidence.'"

  2. LOW Confidence: "Notice the confidence indicator is LOW — the system knows it doesn't have reliable information"

  3. No Hallucination: "A generic AI chatbot would have invented standards. Manak AI refuses to guess — because in compliance, wrong information is worse than no information."

Compliance Gap Analysis Demo (45 seconds — if time permits)

Input Form

Product: Stainless Steel Water Bottle
Material: SS 304
Capacity: 1 litre
Current Tests: Material test, Leakage test
Current Certification: None

What to Highlight

  1. Gap Identification: "The system compares your current status against BIS requirements and shows: Material test ✅ Satisfied, Leakage test ✅ Satisfied, Overall migration test ❌ Not done, BIS certification ❌ Missing"

  2. Actionable Recommendations: "For each gap, it provides specific next steps"

  3. Disclaimer: "Note the clear disclaimer — this is an advisory tool, not official BIS certification"

Closing (30 seconds)

"Manak AI is more than a chatbot. It's a compliance intelligence platform that:

  • Retrieves from authoritative BIS documents
  • Provides clause-level citations
  • Distinguishes mandatory from voluntary
  • Generates compliance pathways
  • Supports Hindi and English
  • Refuses to hallucinate

Our hybrid RAG pipeline combines semantic search, keyword search, and metadata filtering with retrieval-based confidence scoring.

We believe this can be the foundation for a national BIS assistance platform. Thank you!"

Expected Judge Questions & Answers

Q: "How is this different from just using ChatGPT?"

"Three critical differences:

  1. ChatGPT hallucinates BIS standards — ours retrieves from actual documents with citations
  2. ChatGPT can't tell you if certification is mandatory — ours cross-references Quality Control Orders
  3. ChatGPT has no clause-level citations — ours shows exact standard, clause, and page number"

Q: "What if you haven't ingested a particular standard?"

"The system will say it doesn't have sufficient evidence. It never guesses. Additionally, the architecture supports incremental ingestion — new standards can be added by running our ingestion pipeline."

Q: "Is this legally valid?"

"No — and we explicitly disclaim this. Manak AI is an advisory and informational tool. Official certification decisions must come from BIS. Think of it like a knowledgeable advisor who points you to the right documents, not a certifying authority."

Q: "How does it scale to 19,000+ standards?"

"Our pgvector database with HNSW indexing scales sub-linearly. Adding more documents is running the ingestion pipeline. The retrieval latency stays under 200ms regardless of collection size. For production, we'd need more compute but the architecture is the same."

Q: "What's novel / what's your innovation?"

"Two innovations:

  1. The Compliance Pathway — converting raw standard requirements into an actionable step-by-step workflow
  2. The evidence-aware confidence system — retrieval-based, not LLM-generated, so it's grounded in actual data quality"

Q: "Why not fine-tune a model?"

"Fine-tuning bakes knowledge into weights — it can't be updated when standards change and can't cite sources. RAG keeps the knowledge external and citable. For compliance, traceability to source is non-negotiable."

SIH Differentiators / USP

Feature Generic AI Chatbot Manak AI
Citation accuracy ❌ Hallucinates ✅ Clause-level, verified
Mandatory/Voluntary ❌ Cannot determine ✅ QCO-aware logic
Compliance pathway ❌ None ✅ Structured workflow
Confidence scoring ❌ Arbitrary ✅ Retrieval-based
Hindi support ⚠️ Generic ✅ Domain-specific
Knowledge source ❌ Training data (stale) ✅ BIS documents (updatable)
Gap analysis ❌ None ✅ Requirement comparison
Failure handling ❌ Confidently wrong ✅ Explicitly uncertain