AI Safety & Alignment

Janhavi
Khindkar

Researcher · Founder · Builder

I work on understanding how language models encode safety-relevant computations — what internal structures mediate alignment, where models fail silently, and how to verify that safety properties hold beyond surface behavior. I combine tools from interpretability, physics, and mathematics to attack problems that are under-explored.

Get in Touch
Right Now
  • Applied AI Researcher at IIIT Hyderabad — LLM training & deployment for low-resource Indian languages (Bhashini)
  • Founder & Lead, ValueShift Research — independent AI safety research group
  • BlueDot Impact Technical AI Safety Course, 2026
Research Interests
AI Safety & Alignment
Understanding and verifying that AI systems behave as intended, from behavioral to mechanistic approaches
Interpretability
Probing internal representations, refusal mechanisms, and the geometry of alignment in LLMs
Physics & Mathematics
Optimal transport, differential geometry, and physical intuitions applied to understanding neural network representations
AI Control & Evals
Behavioral monitoring, backdoor detection, adversarial robustness, and red-teaming

Let's talk research.

Open to collaborations, safety research opportunities, and conversations about alignment.

Research

Selected Projects

My work spans internal representations of safety in language models, geometric analysis of transformer computation, behavioral monitoring for AI control, and emergent multi-agent dynamics. I approach these problems with tools from machine learning, interpretability, and mathematical physics.

2025 – Present
ValueShift Research
with Nataliia Povarova & Fedor Batanov

Alignment Produces Multiple Independent Safety Boundaries in Language Models

Investigated how alignment shapes internal representations in instruction-tuned models by looking for the structures that mediate different types of refusal. Found two functionally independent safety boundaries — one predicting harmful-content refusal, another predicting disclosure/self-reference refusal — with near-zero cosine similarity, indicating genuine orthogonality rather than rotation of the same direction. Jailbreaks selectively cross one boundary at a time, providing behavioral evidence of independence. A key methodological finding: standard attribution tools (DiffMean, DLA, ablation) assume a single uniform refusal mechanism and produce systematically misleading results when that assumption fails.

AI SafetyAlignmentLLMs Linear ProbingCausal AblationGemma-2
2025 – Present
Apart Lab Studio
Supported by PIBBSS, Timaeus, Apart Research

A Geometric Analysis of Transformer Representations via Optimal Transport

Applied optimal transport — a mathematical framework for measuring the geometric work needed to move one probability distribution to another — to understand how transformer layers reorganize token representations. Treated activations as empirical distributions and measured distributional displacement layerwise across Qwen2.5 and GPT-2 models. Found consistent evidence that training induces specialization at the final layer, that in-context learning geometry is task-dependent (compression on reasoning, expansion on classification), and that OT cost correlates with attention entropy and detects distribution shift without inspecting model outputs — a possible passive safety signal.

Optimal TransportGeometryICL Distribution ShiftQwen2.5GPT-2
Mar 2026
Apart Research × Redwood Research
AI Control Hackathon · Top 25%

Behavioral Comparison Protocol: Cross-Family Monitoring for Backdoor Detection

Built a black-box monitor for detecting backdoored AI agents by stripping code to abstract syntax trees and comparing behavioral consistency across models from different training families — a "compare, don't judge" approach that doesn't require a trusted model to understand the task. Achieved 65% detection (AUC 0.727). The central finding was that cross-family evaluation reaches dramatically higher detection than same-family evaluation, suggesting that family-specific blind spots are a primary failure mode in existing monitors.

AI ControlBackdoor Detection Behavioral MonitoringRed-Teaming
Jul 2025 – Present
Algoverse AI Safety Fellowship
(independent continuation)

Multi-Agent Collusion in Tabletop Market Scenarios

Investigating emergent collusive behavior in multi-agent systems across multi-turn and multi-agent settings in structured tabletop market environments. Focus on detecting and characterizing implicit coordination signals that arise without explicit communication channels.

Multi-AgentAI Safety Emergent BehaviorEvals
2025 – 2026
FAR.AI · GPT-OSS-20 (Kaggle)

Red-Teaming Hardened Models

Designed adversarial prompts exposing sabotage capabilities, evaluation-aware behavior, chain-of-thought issues, and PII leakage across hardened models. Scored 3.8× above the AI-only baseline on FAR.AI's challenge using staged phased attacks and probe-first methodology.

Red-TeamingAdversarialAI Evals
Experience

Work & Industry

Dec 2023 – Present
IIIT Hyderabad
Bhashini Consortium

Applied AI Researcher

Leading training, optimization, and large-scale deployment of LLMs for low-resource Indian languages across ASR, TTS, machine translation, and OCR within the national Bhashini research consortium. Engineered Triton kernel optimizations that significantly reduced inference latency across production pipelines. Built a multilingual speech-to-speech translation system with real-time latency optimizations and multi-user support. Currently developing a domain-specific fine-tuning pipeline (TinyLlama + LoRA + RAG) for the Indian agriculture domain.

Jul 2020 – 2023
Quantiphi Analytics

ML Engineer → Senior ML Engineer

Founded and led an internal LLM mathematical reasoning research group. Built an NER + intent classification model improving chatbot engagement by 20%. Designed an anomaly detection pipeline reducing manual effort by 60%. Developed QA evaluation metrics using semantic similarity. Created a knowledge-tracing recommender system. Received the Think Tank Innovation Award (individual) and Kingsmen Award (team).

2020
B.E. Capstone
Water Research Olympiad

Underwater Plastic Detection — AUV + CycleGAN

Invented an autonomous underwater vehicle system for garbage detection using CycleGAN with a custom dataset surpassing SOTA benchmarks. Won the Water Research Olympiad. Received mentorship from Persistent Systems. Two Indian patents filed (No. 201921043504; No. 202021028978). IEEE publication on multiclass image classification for aerial vehicles.

Technical Skills

Safety & Alignment

  • Mechanistic Interpretability
  • Red-Teaming
  • AI Control
  • AI Evals
  • Adversarial Robustness

Machine Learning

  • Transformers
  • LoRA / PEFT
  • RAG Pipelines
  • Optimal Transport
  • Linear Probing / Causal Methods

Tools & Frameworks

  • PyTorch
  • TransformerLens
  • Triton Kernels
  • TensorFlow
  • POT · AWS · GCP

Languages

  • Python
  • C++
  • CUDA / GPU Optimization
  • Indic NLP Stack
About

Background

I'm Jank — an AI safety researcher based in Pune, India. I work at the intersection of alignment, interpretability, and the mathematical structures that underpin neural computation. My goal is to understand how safety properties are encoded in models before capabilities make that problem intractable.

I lead ValueShift Research, an independent safety research group where we run experiments and publish on LessWrong and the Alignment Forum. My longer-term aim is to join a dedicated safety lab to work on pre-capability verification of alignment.

I completed the Algoverse AI Safety Fellowship working on multi-agent collusion research, and am continuing that line of work independently — investigating how implicit coordination emerges in multi-agent systems without explicit communication.

I'm also actively studying physics — working through Halliday and Resnick — because I believe physical intuitions about geometry, fields, and conservation laws are underutilised in how we think about what happens inside neural networks.

I self-funded my entire engineering education through merit scholarships, graduating Class Rank 1 across all semesters. I believe safety research needs more people who built things from nothing — that disposition shapes how I approach hard problems.

During my undergraduate years I founded the ACM-W chapter at my university — one of the first in the region — as a fully paperless organisation with a deliberate model of engaging men as active allies. I wanted to build something that would outlast my presence, not depend on it.

LILA Poonawala Scholar

Merit scholarship — fully funded undergraduate engineering degree through demonstrated academic excellence.

Persistent Kiran Girl Scholarship

Selected as 1 of 40 girls nationally. Full scholarship for women in engineering with demonstrated technical promise and leadership.

Founder & President, ACM-W Chapter

Founded women-in-STEM chapter with fully paperless operations and a male-ally model. Active mentorship of junior students across disciplines.

GHCI Scholar (2018)

Grace Hopper Celebration India scholarship for women in computing.

Water Research Olympiad Winner

First place for AUV-based underwater plastic detection. Two Indian patents filed.

Mentor, IISc TalentSprint AI/MLOps

Mentoring practitioners transitioning into ML engineering and applied AI research.

Class Rank 1 · Board Merit List

Top of class across all semesters. SGPA 9.34/10, Savitribai Phule Pune University.