Meet Auto-Discovery, now live in Novix. It turns every research question you've explored into a research constellation, then reads it to surface the next direction worth pursuing. It proposes. You decide: go ahead, or not now. Your research should compound. Now it moves forward on its own.

Research question: how do crystal symmetry, elastic stability, and directional mechanical response compare across real elastic tensors? Novix ran the computational reanalysis and produced the following: ▪️ recovered and checksum-verified the historical Materials Project elasticity dataset ▪️ audited 1,181 tensors across six crystal systems ▪️ recomputed Voigt-Reuss-Hill moduli, universal anisotropy, directional Young’s modulus, shear envelopes, and Poisson-ratio envelopes ▪️ quantified structure-property associations with formula-aware and crystal-system sensitivity checks ▪️ compared two carbon records with sharply different local geometry and directional stiffness ▪️ generated a 33-page paper with 14 figures, tables, equations, references, and an executable analysis package

Give Novix a paper. It reads the method, rebuilds the pipeline, and runs the experiment. Reproduction is where a claim becomes evidence.

Got a half-finished paper you’ve been avoiding? Try this in Novix: → In Auto Pilot, ask Novix to read your draft like a reviewer. Let it point out weak arguments, unclear claims, and the places a reviewer would likely get confused. Basically, the problems you stop seeing after staring at your own draft for too long. → Then step back and review the paper yourself. Use Novix’s feedback as a starting point. Reread the paper as a reviewer, not as its author. That shift in perspective can make many of the problems easier to untangle. → Use Annotation for targeted edits. Screenshot the exact part you want to change and tell Novix what you want: delete it, rewrite it, go deeper, or add another angle. Get the core into shape first, then keep polishing. Most of the resistance only shows up at the beginning. Once you sit down and take the first step, the work gets less scary, and you start finding that natural state of actually doing it.

Research papers are becoming executable. A Nature paper, Paper2Agent, turns a paper into an agent you can query, rerun, and connect to other papers. The next step isn’t chatting with a PDF. It’s letting a paper apply its methods to new data, check its claims, and collaborate with other agents. A paper-specific agent is useful. A research orchestrator connecting literature, hypotheses, code, experiments, and writing is the bigger shift. That’s the direction we’re building at Novix.

Novix Workflow Tip #01

How to write a literature review draft in 1 hour？ 1. Open novix.science 2. Go to Deep Survey 3. Enter your review topic 4. Novix asks you questions first to clarify the task 5. Novix then writes a review draft for you

Research question: which weather signals actually drive photovoltaic power, and which model forecasts it best? Novix ran this question and produced the following: ▪️ built a reproducible benchmark for hourly German aggregate PV forecasting on public data (OPSD generation + ERA5 weather) ▪️ engineered the weather and calendar features: shortwave radiation, humidity, temperature, cloud cover, lagged capacity factor, calendar encodings ▪️ trained and compared four models: OLS linear regression, random forest, XGBoost, and LSTM ▪️ evaluated on a strict chronological split: train on 2017, select on 2018, test on a held-out 2019 (4,043 daylight hours) ▪️ quantified significance with daily block bootstrap intervals and paired daily-loss tests ▪️ opened the black box with TreeSHAP, ALE, and partial dependence to see which signals move predictions ▪️ wrote up a full paper draft with figures, tables, equations, and references

Most lit reviews die in the tab graveyard. 40 open PDFs, 3 highlighted, 0 synthesized. Novix takes 100 at a time, maps where they disagree, and shows you which gap is still unclaimed. A review is worth writing when it ends with a question.

Novix now runs on stronger models. Here’s what’s improved: 1. Complex experiments run more autonomously, more reliable long-running tasks 2. Deeper reasoning, stronger originality and novelty 3. Stronger large-scale data processing, model training, analysis, and visualization 4. Higher-quality outputs and better academic writing Live now for every user.

How to spend your first PhD semester well: Treat it like due diligence. Test 2–3 research directions with small projects. Do one real piece of work with your advisor and learn how they actually give feedback. Then pick one direction and go deep enough to see the ugly parts: failed experiments, dirty data, long waits. Meanwhile, figure out what your field is arguing about: conflicting results, metric debates, competing explanations. By the end of the semester, write a 5-page research plan answering two questions: Which problem do you think is worth pursuing? What will you do if your core assumptions fail? The most important thing in your first PhD semester is to rule out, at the lowest possible cost, a few choices that could otherwise waste the next five years.

How to lock down a candidate research gap in 2 hours： 1. Compress a broad interest into searchable questions. (0-15 min) 2. Build a keyword matrix to expand your search space. (15-40 min) 3. Extract gap evidence from abstracts and discussions. (40-70 min) 4. Decide whether it is actually a gap. (70-90 min) 5. Turn the gap into an academic research question. (90-105 min) 6. Create a research gap action map. (105-120 min)

Five habits for developing stronger thesis ideas: 1. Before asking what hasn’t been done, ask why it hasn’t been done. A gap in the literature may be an opportunity, but it may also be a warning sign. 2. Come up with three serious objections to each direction. If you can’t, the idea probably hasn’t moved beyond a promising title. 3. Never develop just one idea. You can only decide what is worth pursuing when you have alternatives to compare. 4. Use a repeatable process instead of starting from a blank page. Novix’s Idea Spark turns a research field into an idea tree, showing the references and limitations of each branch. You’ll rule out most branches. That process of elimination is where much of the value lies. 5. Take three directions to your advisor. Bringing one invites a verdict. Bringing three starts a conversation.

👉novix.science

How to keep going when research gives you no feedback? Don't turn a lack of feedback into negative feedback. Solve what's in front of you. Rest when needed. Continue tomorrow.

How to avoid getting stuck in the details when starting a research project? Build the structure first, then fill in the details. Take writing a paper as an example: 1. Outline the sections and define the question each one answers. 2. Under each section, specify the deliverables: how many cases, how many figures, what to analyze, and which metrics to use. 3. Let the first draft be rough. Mark missing pieces as TODOs and fill them in as you go. This is also where tools like Novix can help: getting the first pass down quickly. The key is to externalize your thinking. Details that seem complicated in your head usually get simpler once written down. With the structure in place, you just work on the next box instead of deciding from scratch every morning.

A good research paper introduction is like a funnel: it starts broad and gradually narrows down. Why the field matters. What's already been done. What's still missing. What you do about it. Each part should naturally set up the next.