(The way we’ve always done it. The way they live it.)
# Human Intelligence
- Conclusion: We need a better way.
- Limitations: Time. Bias. Scale. Burnout.
- Results: 1 report. 1 officer. 1 deadline missed.
- Methods: PubMed → Embase → 10,000 abstracts → 3 months
- Background: What we know, what we don’t
“This is your day. This is your burden. This is what we’re here to change.”
(Not magic. Not replacement. Augmentation—by type.)
# AI Powered
- Five types of AI → Five types of tools
5. Embodied AI → Ultimate goal (no human input needed at any point, Full automation)
4. Generative AI → Disciplined branches (solving problems in data collection, analysis, reporting, ethics)
3. Agentic AI → Pulsing trunk (speed of delivery of results, Human-in-loop)
2. Perception AI → Smarter roots (rigorous analysis)
1. World AI → Faster soil (data access, Hybrid intelligence)
“We don’t replace you. We grow with you. One layer at a time.”
(Not global. Not generic. Theirs.)
# WHO-India
- Conclusions: Helpful to have precise language
→ Vocabulary beyond “AI tools”
→ Understand features of various types of AI
→ The industry's goal is a fully-stack embodied AI (the only form AGI can take)
- Limitations: Invariant aspects of science & public health
→ Bias must be named (by experts, human standards)
→ Gaps must be acted on (collecting, finding, accessing data)
→ Evidence must be found (knowledge gaps)
- Results: Tailored to WHO-India
→ policy speed
→ PRISMA, equity
→ 700+ officers
- Journal & language agnostic
→ Hindi, Tamil, etc (eg OpenEvidence only accesses NEJM, JAMA: Journals in English)
→ Grey lit
→ Field notes
- Literature most relevant to India
→ NCDs
→ TB
→ AYUSH, UHC
“This isn’t AI for AI’s sake.
This is your evidence, your language, your impact—made faster, fairer, scalable.”