How I Review 200 Research Papers as Dean — My AI Workflow

Last Tuesday, I sat down at 6:30 in the morning with a cup of chai and opened my inbox to find 47 unread emails — 12 of them containing research papers from my PhD scholars, 8 from faculty across our five departments, and the rest administrative threads that could not wait. By 9 AM, I had a faculty meeting. By 11 AM, a curriculum review session. And somewhere between all of that, I was supposed to provide meaningful, substantive feedback on research work that people had spent months producing.

This is not a complaint. This is the reality of being Dean of the School of Sciences at Woxsen University while also mentoring doctoral researchers, running collaborative projects, and staying current in my own field. If you are a research faculty member or a PhD scholar reading this, I suspect you recognise the feeling — the weight of intellectual responsibility pressing against the hard limits of time.

So let me tell you exactly what I changed, and how I now review more than 200 research papers every academic cycle without sacrificing depth or rigour.

The Problem Nobody Talks About at Dean Level

When I completed my PhD in Nuclear Thermal Hydraulics and CFD at IIT Guwahati, my world was beautifully narrow. I had one problem, one supervisor, one set of journals, and all the time in the world to go deep. I read papers slowly, annotated margins obsessively, and rebuilt every simulation from scratch to understand it.

That version of me would not survive a single week in my current role.

At Woxsen, I oversee five departments — Physics, Chemistry, Mathematics, Data Science, and Biological Sciences. We have more than 500 students currently enrolled, active PhD programmes, and a research culture that I am personally invested in building. I also collaborate with our Care and Cure clinic, where the intersection of computational biology and clinical practice generates a constant stream of research questions that require my attention and input.

The honest number is this: in a given semester, I am expected to meaningfully engage with upwards of 200 research documents — this includes journal manuscripts submitted by faculty for pre-submission review, PhD progress reports, thesis chapters, conference papers, literature surveys from scholars, and collaborative grant proposals. Two hundred is not an exaggeration. It is, if anything, a conservative count.

For years, I managed this the way most academics do — by working harder. Earlier mornings. Later nights. Weekends sacrificed. The quality of my feedback suffered even when the quantity looked fine on paper. I was skimming when I should have been reading. I was pattern-matching from memory when I should have been engaging with the actual argument in front of me.

Something had to change.

What I Actually Do Now — The Workflow in Detail

About eighteen months ago, I began systematically integrating AI-assisted tools into my research review process. I want to be precise about what this means, because there is a great deal of noise around "AI in academia" that conflates very different things.

I am not using AI to read papers for me. I am using it to read with me — faster, more systematically, and with fewer cognitive gaps caused by fatigue or time pressure.

Here is the actual sequence I follow.

When a paper lands in my review queue, the first thing I do is upload the document and ask the AI to generate a structured summary: the central claim, the methodology, the key findings, and the limitations the authors acknowledge. This takes less than two minutes and gives me what I call a "research scaffold" — a skeleton I can verify and interrogate rather than construct from scratch.

I then read the paper myself, but I read it differently. Instead of reading linearly from abstract to conclusion, I use the scaffold to navigate. I go directly to the sections where the scaffold summary seems uncertain or where I have domain-specific doubts. For a CFD paper, I will immediately check the boundary condition assumptions and mesh independence study — because I have run enough simulations at IIT Guwahati to know exactly where the soft spots tend to be. The AI cannot replace that instinct. But it frees me to apply that instinct selectively rather than exhausting myself on every page.

The second stage is what I call gap interrogation. I ask the AI to compare the paper's claims against its cited literature — specifically, to flag whether the authors have cited work that contradicts their conclusions and addressed it, or whether there are conspicuous silences in the literature review. For our PhD scholars working on interdisciplinary topics at the care and cure interface — say, computational modelling of drug delivery or biosensor signal processing — this cross-domain literature check catches errors that even experienced reviewers miss because they are specialists in only one of the relevant fields.

The third stage is feedback generation. After I have formed my own views, I dictate my comments — sometimes as voice notes, sometimes as rough bullet points — and use the AI to help me structure them into reviewer-quality prose. This matters more than it sounds. When a junior PhD scholar receives vague feedback like "the methodology needs strengthening," they often do not know where to begin. When the same observation is written as "the turbulence model selection is not justified relative to the Reynolds number regime under study — consider comparing k-epsilon and k-omega SST results and discussing the choice in Section 3.2," it becomes actionable. The AI helps me maintain that level of specificity across every paper I review, not just the ones I happen to review on a good day.

What This Has Actually Changed

The numbers are real. My average review turnaround has dropped from eleven days to four days. The length and specificity of my feedback has increased. More importantly, the scholars themselves have told me — in our research group meetings and in one-on-one supervisions — that the feedback feels more engaged, not less. That is the metric I care about.

For faculty across our departments who are preparing manuscripts for submission, I have introduced a pre-submission review protocol that runs through the same AI-assisted pipeline. A Chemistry faculty member submitting to a Q1 journal now receives a structured critique addressing significance, novelty, methodological rigour, and presentation — in under a week, consistently, rather than in three weeks when I can finally find the time.

I have also found unexpected benefits in my own research. Staying current with literature in Nuclear Thermal Hydraulics while simultaneously advising work in biological sciences and data science was becoming impossible. The AI-assisted literature monitoring I now run means I can maintain genuine awareness across fields without the cognitive cost of reading every paper fully before deciding whether it is relevant.

The Principles Underneath the Workflow

I want to offer three principles that I think are transferable, regardless of what tools you use.

First: AI is a force multiplier for expertise, not a replacement for it. The reason my workflow produces good feedback is that I bring thirty years of research training to every interaction with the tool. Without domain depth, the output is shallow. With it, the output is fast and substantive.

Second: the goal is better scholarship, not faster throughput. I am not trying to review 200 papers in 200 minutes. I am trying to ensure that every researcher in my school receives feedback that genuinely improves their work. Speed is a by-product of the workflow, not its purpose.

Third: build the habit before you build the system. I spent six months experimenting before I settled on a repeatable process. There is no shortcut to that learning period. Every discipline, every research culture, and every individual reviewer will need to find their own calibration.

A Final Word for PhD Scholars

If you are a doctoral researcher reading this, I want to say something directly: the same workflow that helps me review your work can help you produce better work in the first place. Running your own draft through a structured AI review before you submit it to your supervisor is not cheating. It is the same thing as asking a peer to read your work — except the peer is available at midnight when your submission deadline is tomorrow morning.

The scholars in my programme who have adopted this practice are submitting cleaner drafts, having more productive supervision meetings, and progressing faster through their PhD milestones. The research is still theirs. The thinking is still theirs. The AI is simply helping them see their own work more clearly before anyone else does.

That, ultimately, is what good tools do. They do not think for us. They help us think better.