As a Program Director at Algoverse, I oversee AI Research programs and help grow global AI safety initiatives through fellowships and community-building.
I am the Research Program Director at PRISM (Peer-vetted Research Initiative for Safety Methodologies), a 16-week AI safety research program run under Women Who Do Data and funded by the Long-Term Future Fund (LTFF). PRISM uses peer selection to scale mentorship and turn the senior-researcher bottleneck into a multiplier.
When I'm not thinking about AI, you'll probably find me cheering on indie game creators as a Community Director at the International Game Developers Association (IGDA).
I work on AI safety, interpretability and evaluation, most of it co-authored with the early-career researchers I mentor at Algoverse and PRISM. Every submission, review and decision below is public on OpenReview.
A five-seed controlled study of whether a layer-redundancy regularizer can stop QA models from contradicting themselves. An honest negative result.
Prediction markets only work when beliefs are independent. We show preference-optimised LLM traders converge into a correlated monoculture.
Reads latent chain-of-thought methods such as CODI through a dynamical-systems lens to explain how continuous reasoning trajectories actually evolve.
A black-box audit of LLM oversight layers on a grid-control agent: 0 of 30 explanations flagged an adversarial 600 MW observation injection.
Moral reasoning across English, Spanish, Korean and Mandarin in five LLMs. Cultural grounding collapses by up to 88% outside English.
Frontier judges happily validate reasoning chains where a third of the steps are plausible-sounding filler. Our verifier cuts false positives from 37% to 3%.
The “Gaslight Protocol”: vision-language models asked to judge copyright similarity follow the prompt over the pixels, hallucinating forensic evidence.
A benchmark of realistic, messy human queries for measuring how well agents retrieve the right Model Context Protocol tool.
Market making as a scalable coordination mechanism for keeping multi-agent LLM systems calibrated and aligned.
Our MEDIQA shared-task system for medical natural language inference, with attention visualisation for clinical interpretability.
Selected by The Infocomm Media Development Authority (IMDA) among 800 nominees
Read More
Panelist for the Women in Data Science (WiDS) Geneva conference into the real challenges data science teams are facing right now.
Read More
Judge for Outstanding Outstanding Citizen Contributor Award
Read MoreSix years of talks, panels and keynotes — scroll across the timeline.
Keynote at Google Developer Group Bandung on how to quantize models for Tensor Processing Units.
Walked through building a ChatGPT-powered Chrome extension that converts informal writing into a more professional form.
Saama Connect meetup talk on how we can ensure AI systems remain safe and beneficial as they become more powerful.
Panelist at Women in Data Science (WiDS) Geneva on the real challenges data science teams face today.
GDG Dublin International Women's Day keynote on why AI-First Architecture could be our next big engineering crisis.
Demonstrated Scaledown, a free tool that helps users track the carbon footprint of their AI prompts.
I've designed and instructed these courses on LinkedIn Learning, teaching practitioners how to ship machine learning, generative AI, and LLMOps systems in production.
Author & Instructor
LinkedIn Learning
Author & Instructor
LinkedIn Learning
Author & Instructor
LinkedIn Learning
Author & Instructor
LinkedIn Learning
Stats, ratings, and links →
Discover the impact of Archana's expertise through the voices of peers and professionals who have witnessed her transformative influence in AI, ML, and tech diversity
President, Women Who Code
Archana is an inspirational business and community leader who offers a unique blend of technical and business skills. I am continuously impressed by her ability to glean thoughtful insights through data, think outside the box to solve challenges, and work cross-functionally with teams across the organization to support big-picture thinking. She is genuinely delightful to work with, full of energy, and always prepared for a challenge. Highly recommend!
Board Advisor and Founder
We have worked alongside each other as Fellows at Women Who Code for some time, but this was our first opportunity to really build and execute a project together. I was particularly impressed by Archana's high level of technical proficiency, and her depth of technical knowledge. Further, her teaching style invites learners in to engage and ask questions. No matter the time of day, Archana shows up with a smile and a huge bucket of enthusiasm. Archana is certainly an asset to any team lucky enough to have her.
Senior Software Engineer
Archana is very thoughtful and hardworking. I had the pleasure of working alongside her at Women Who Code and I admired her keen ability to analyze, research, and solve problems - as well as provide excellent communication and documentation along the way. She is very insightful about product and system needs, while also being considerate and supportive of her colleagues. Archana would be an incredible asset to any team.
Associate Director - Recruitment
Have known Archana for quite some time and she sparks off as a very postive and engaging individual. What impresses me about Archana is her love for the Technology/Women in Technology community where she volunteers and helps out at a lot of external events on top of her day job. She has a lot of passion for Technology and is inspiring in the ways that she goes above and beyond. She would be a lovely addition to any firms she joins and any firm would be lucky to have her!