Personalized research feed

The agent keeps a standing watch on the literature for you. Instead of you re-running the same searches every week, it learns what you care about from the papers you save and quietly surfaces new work that matches — catching relevant results even when the authors describe them in words you would never have thought to query.

#Watching the literature for you

Your personalized feed is a continuous intelligence stream over 40M+ publications. The agent monitors newly published work across the life-science literature and scores each candidate against a model of your research interests, then presents the closest matches ranked by how well they fit your context. It is the difference between a static keyword alert that fires on an exact phrase and an assistant that understands the meaning of what you study.

The watch runs on a few consistent rules:

• A recent window. The agent evaluates papers published in roughly the last three months, so the feed always reflects the current leading edge rather than re-surfacing work you have already seen.
• Meaning over keywords. Matching is done on semantic similarity, not literal string overlap, so the feed catches relevant papers that use different terminology for the same genes, pathways, methods, or mechanisms.
• Always on. You do not schedule or re-trigger the watch. As new literature is indexed, the agent re-scores it against your profile and updates the feed on its own.

#Building your semantic profile

The agent cannot read your mind, so it reads your library. Every paper you save into a library folder is a signal about what matters to you, and the collection of those saved papers becomes the raw material for a semantic profile — a learned representation of your interests expressed in the same embedding space the agent uses to compare papers.

The profile is built and maintained as follows:

1. You save papers. During normal research sessions you add papers to library folders. These define, by example, the topics and sub-topics you work on.
2. The agent learns an embedding. Using contrastive-learning embeddings trained on your saved collection, it distills the breadth of your library into a compact profile of your interests rather than a single average topic.
3. The profile stays current. As you save more papers — and as you rate feed recommendations — the agent refines the profile so it tracks how your focus shifts over time.

Because the profile is learned from meaning, it generalizes. A library centered on one disease context can still surface papers describing the same underlying biology in a different system, which is exactly the kind of connection a keyword alert would miss.

#How recommendations are ranked

Once the profile exists, ranking the feed is a scoring problem. Each new paper in the recent window is compared against your semantic profile, and its similarity determines where — and whether — it appears in your feed.

• Relevance scoring. The agent computes how closely each candidate paper matches your profile in the embedding space, and orders the feed so the strongest matches sit at the top.
• Conceptual, not lexical. Because scoring is semantic, a paper can rank highly without sharing any of your typical search terms — it earns its place by discussing the same concepts, mechanisms, or techniques.
• Explained matches. Each recommendation carries an agent-written note on why it was surfaced for you specifically, so you can judge relevance without opening every abstract.

This is what separates the feed from a synonym list: it does not just widen your keywords, it reasons about closeness in meaning, which is why it can act as an early-warning system for adjacent breakthroughs that connect to your field.

#Semantic matching versus keyword alerts

Traditional keyword alerts fire only when an exact term appears, so they silently miss papers that describe the same idea in other words. The agent's semantic matching closes that gap in three concrete ways:

#Digests and refining results

The feed is a two-way channel. The agent brings you papers, and your responses teach it to bring better ones — while email digests make sure high-relevance matches reach you even when you are not actively browsing.

#Staying informed

• Email alerts. When new high-relevance papers match your profile, the agent can notify you by email so you never have to check the feed manually to stay current.
• Pull into research. Any recommended paper can be imported straight into an active research session, so a promising match flows directly into deeper analysis.

#Refining what you see

• Rate relevance. Mark recommendations as relevant or not relevant, and the agent folds that feedback back into your profile to sharpen future ranking.
• Dismiss the misses. Dismissing an off-target paper is a signal too — it tells the agent what to down-weight, so the feed drifts steadily toward your real interests.

Over time this loop compounds: the more consistently you rate and dismiss, the more the profile reflects your genuine priorities rather than a one-time snapshot of your library.

#Tips for a better feed

A little curation goes a long way. These habits give the agent the clearest possible picture of what you want to be told about.

• Keep saving papers. The profile is only as rich as your library. Add papers as you read so the agent learns from a growing, current collection.
• Cover every facet. Include papers from the different aspects of your work so the agent captures your full breadth, not just your most recent topic.
• Rate consistently. Even occasional relevant / not-relevant feedback measurably tightens ranking — treat it as training, not busywork.
• Prune deliberately. Dismiss papers that miss the mark rather than ignoring them, so the agent has an explicit negative signal to learn from.