Foresight
About
About

Every paper says what
should come next.

Foresight reads the future-work sections of 1,600+ ML papers, finds the patterns across them, and synthesises what researchers are collectively asking for. Then it lets you implement those predictions and run them — right here.

How It Works

01

Explore

Papers from NeurIPS, ICLR, ICML, CVPR, and arXiv — connected by citations and semantic similarity. Browse any field as an interactive timeline and see how papers relate to each other.

02

Predict

Most papers end with a future-work section — what the authors would build next. We extract these from 700+ papers and synthesise the patterns into predicted research directions, each grounded in specific source papers.

03

Implement

Click any prediction and a Claude agent reads the source papers, writes a working Python prototype, and runs it in a sandbox. You go from a predicted research direction to executing code in under a minute.

By the Numbers

1,663
papers indexed
2,266
citation edges
35,384
semantic connections
19
future directions
726
future work sections extracted
70
researchers matched

Architecture

Data Pipeline: Semantic Scholar API → Supabase (Postgres + pgvector) → Next.js 15
AI Pipeline: Claude Agent SDK (streaming SSE) → E2B Sandbox (live Python execution)
Frontend: D3-force graph · Custom SVG · Tailwind CSS · Crimson Pro + JetBrains Mono
Next.js 15TypeScriptTailwind CSSSupabaseClaude SDKE2BD3.jsVercel

Bounty Tracks

Anthropic

Best Use of Agentic SDK

Claude does three jobs here: synthesising future-work sections into predictions, generating implementations that reference source paper methods, and writing personalised research briefs. Code generation streams live via SSE — the user watches the agent read, reason, and write.

Lovable

Best UI

Designed to feel like a research tool, not an app. Serif type, monochromatic palette, and a D3-force timeline you can zoom, pan, and click through. Papers cluster by similarity. Future directions pulse. Click a node and see exactly why it exists.

Team

Will Rowan

  • PhD Computer Science, University of York. Thesis on 3D face reconstruction
  • Co-founder of PXLD, an AI video relighting startup (featured in Variety, Screen Daily)
  • Published at ICLR, ECCV. Best Poster Award, ECCV 2024
  • Research Fellow working on AI-powered virtual production tools
w-rowan.github.io

Simon Luo

  • BA Computer Science, Churchill College, Cambridge - thesis on applying ML models to financial market prediction
  • Former Technical Lead and FDE at Palantir
  • Former SWE at DRW
  • Currently exploring startup ideas worth building
linkedin.com/in/luo-simon