Tanish Anandababu
Software engineer. Researcher.
Between disciplines, on purpose.
I study computer science at Maryland, with minors in advanced cybersecurity and computational finance. Adversaries and capital are two of the places real systems get tested, and the minors are how I explore both.
Six roles so far - Google, Ernst & Young, U.S. News, a computational biology lab at UMD, a maritime operations startup in Mumbai, and a first-author ACM paper on LLMs in mathematics education.
I'm drawn to projects where software has to answer to something outside itself, whether a car, a ship, a drug candidate, or a roomful of stakeholders. Cross-disciplinary work is what fascinates me.
Computational biology, cybersecurity consulting, maritime operations. Different constraints, different goals, different rules of the room. The skill I lean on most is learning those rules quickly and adapting my engineering to fit.
On-device, on the road.
Joining the Android Auto team to develop on-device AI-powered features for the Android Auto ecosystem - a surface where latency, safety, and ambient context all arrive at once and none of them can be traded off.
On-device AI is where the next decade of ambient computing gets decided. The car is the hardest surface to get right - no margin for latency, no tolerance for hallucination, constant environmental context. Incoming.
An agent that reviews the reviewers.
Built an end-to-end agentic AI pipeline for the cybersecurity practice - a knowledge base for secure document retrieval paired with a critique agent that returned actionable feedback and revision suggestions on cybersecurity documentation.
Designed for reuse, the pattern scales across teams and engagements. Demoed directly to the external client and to senior managers and partners inside the practice.
An agentic system is only as useful as the trust it commands. The technical architecture was complex, but the real bottleneck was the review loop. The software had to answer directly to senior practitioners and external clients, proving it could critique securely and accurately.
Listening, then shipping.
At U.S. News - engineered a data-validation and QA platform over the internal API, leveraging advanced heuristics and data analysis to flag anomalies and tighten editorial quality. Designed and executed API optimizations that reduced response times by over 50% for endpoints handling large data extractions. Built React interfaces with Flask and Jupyter pipelines, and long-horizon metric visualizations to inform decision-making.
At Volteo - shipped the Wayship vessel operations platform's most-requested dark mode end-to-end, grounded in user research with ship captains. 40% user satisfaction lift, 60% more nighttime usage.
Product engineering is mostly listening. Volteo's dark mode came from talking to ship captains. The U.S. News QA platform came from understanding what editorial teams were already flagging by hand. The code is the second half.
From membranes to models.
Research under Dr. Jeffery Klauda at the Laboratory of Molecular and Thermodynamic Modeling. Studying how ligands alter the conformational dynamics of EAG1, a voltage-gated potassium channel overexpressed in several cancers. Built a config-driven Python pipeline on UMD's Zaratan HPC cluster that processes MD frames across several simulations, using dihedral-angle clustering and PCA. Collaborating with postdoctoral researchers at Georgetown Medical Center on EAG1 as a live anticancer drug target.
At FIRE - first author of “Evaluating Large Language Models in Undergraduate Mathematics: Balancing Potentials and Pitfalls,” accepted by the ACM Journal of Responsible Computing.
Research teaches a discipline I can't get from shipping alone: the precision of a claim. Whether it's membrane dynamics on HPC or LLM accuracy on undergraduate proofs, the question is always the same - what exactly does the evidence let me say?
The stack is wide
because the problems are.
I don't collect tools; I pick them up when a problem asks for one. Every language on this list is here because a project needed it. When the next project asks for something I haven't used, I'll learn that too.