A nonprofit civic interface that uses AI to translate active Indian legislation into plain language via a dual-framing system — studying how interface design shapes civic comprehension and epistemic trust.
Check it out nowAbout
I am an independent AI safety researcher based in Chennai, working on what fine-tuning does to a model’s internal structure and behaviour.
I came to this through CitizenView, a civic-tech platform I built that renders Indian legislation into plain language for roughly 400 users — marketing it surfaced real concerns about AI in sensitive civic work, which pushed me toward the AI Safety Fundamentals fellowship at MIT AI Alignment, the Nous AI Safety x philosophy reading group, and running a REINFORCE experiment on gradient starvation.
I’m now working through ARENA, measuring how narrowly LoRA adapters localise behavioural change, with model diffing as my route into mechanistic interpretability.
Projects
Independently arrived at the hypothesis that sigmoid scaling curves and apparent capability ceilings are measurement artifacts — built a simulation separating reward smoothing from difficulty spread, then found this aligns with established work (Schaeffer et al., 2023) showing emergent abilities are a mirage of metric choice.
View on GitHubEducational visualizer that maps mental health disorders across three connected levels — brain regions, drug mechanisms, and gene pathways.
View on GitHubRecognition
Program
CitizenView accepted into the Sarvam Startup Program
Sarvam AI
Award
Best Participant
Youth Internet Governance Forum
Panel
Co-chair — “Cancel Culture: Censorship vs Perspective”
Mount Carmel College · Panel on contemporary digital expression
Experience
Fellowships
Better Futures Fellow
Aug 2026 – Present · Part-time · Remote
Selected for a nine-week pilot fellowship on how advanced AI could shape humanity’s long-term future — intelligence explosion, power concentration, collective epistemics, and desirable long-term futures, drawing on Forethought’s Better Futures research programme, culminating in a published research or field-building project.
Massachusetts Institute of Technology AI Alignment Club (MAIA)
AI Safety Fundamentals (AISF) Fellow
Jun 2026 – Present · Remote
Building technical foundations in machine learning as groundwork for alignment research.
Nous AI Safety × Philosophy Reading Group
Mentee
Jul 2026 – Present · Remote
Discussing the philosophical assumptions underlying AI safety — consciousness, alignment, agency, and governance
Youth Internet Governance Forum India
Fellow
Aug 2025 – Present · Online / Goa
Youth perspectives in national internet governance policy dialogues
India School on Internet Governance (inSIG)
Fellow
Aug – Sep 2025 · Goa
AI governance, cybersecurity & multi-stakeholder policy design
Certifications & Education
McKinsey Forward Program
Certified Participant
Dec 2025 · Online
Business strategy, structured problem solving & data-driven decisions
Hertie School Data Science Lab
Data Science Summer School
Aug 2025 · Online
Python, data science & analytical methods at a leading governance school
Jyoti Nivas College
BA Journalism, Media Studies & English Literature
Aug 2022 – Apr 2025 · Bengaluru
Film Circle, InkPrint · screenwriter & assistant director
Work
The Parallel Cinema Club
Branding & Strategy
Jul 2024 – Mar 2026 · Bengaluru
Grew reach to 77.3% non-followers · multi-platform content & analytics
Nüpop Records
AD & PR Intern
Mar 2025 · Bengaluru
30-piece content strategy from trend analysis
Editorial Board of JNC
Editing Head
Mar 2024 – Apr 2025 · Bengaluru
Apprentice → Head in 4 months · 11 published articles
Together Magazine
Content Writing Intern
Jun 2023 – Jul 2023 · Bengaluru
Articles on policy, gender parity & biohacking
Internet Society
Member
Oct 2025 – Present · Online / Mumbai
Newsletter content & internet governance networking
Research
Commissioned Research
SRHR Stakeholder Analysis
Policy Researcher · Oct 2025
Influence mapping across 20 key political actors on sexual & reproductive health rights in India — analyzing parliamentary discourse, judicial flashpoints, and advocacy terrain.