The latest articles on DEV Community by Flynn L (@flynn_l_e173fef9d85a8b4a5). https://dev.to/flynn_l_e173fef9d85a8b4a5 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3789791%2F7a4b0924-4486-448d-a2b6-12f8231a974c.png https://dev.to/flynn_l_e173fef9d85a8b4a5 en Flynn L Tue, 03 Mar 2026 15:51:10 +0000 https://dev.to/flynn_l_e173fef9d85a8b4a5/show-dev-i-built-an-ai-research-workbench-with-react-fastapi-and-multi-agent-llms-46fm https://dev.to/flynn_l_e173fef9d85a8b4a5/show-dev-i-built-an-ai-research-workbench-with-react-fastapi-and-multi-agent-llms-46fm <p>I built Arbor, a tool that takes research questions and builds interactive knowledge graphs from academic papers.</p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F57hlzw76kr40wys9i578.gif" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F57hlzw76kr40wys9i578.gif" alt=" " width="760" height="413"></a></p> <p>How the pipeline works</p> <ol> <li>Moderation — GPT-4o-mini screens the question for prohibited content</li> <li>Decomposition — Gemini 2.5 Flash breaks the question into 4 focused sub-inquiries</li> <li>Search — Parallel queries to arXiv and Semantic Scholar APIs</li> <li>Screening — Gemini 2.0 Flash scores papers for relevance</li> <li>Extraction — Gemini 2.0 Flash pulls key findings from top papers</li> <li>Synthesis — Gemini 2.5 Flash assembles findings into structured markdown</li> </ol> <p>All stages stream via SSE, so the graph builds in real-time.</p> <p>One interesting architecture decision:</p> <p>I used a fallback decomposition strategy. If the LLM decompose call fails (timeout, API down, rate limited), the backend generates 4 templated inquiries based on the question stem. The research continues with degraded but functional results instead of showing an error. Users barely notice.</p> <p>Stack</p> <ul> <li>Frontend: React 19, TypeScript, Vite, Tailwind, React Flow</li> <li>Backend: FastAPI, SQLAlchemy, aiosqlite</li> <li>LLMs: OpenRouter (Gemini 2.5 Flash, Gemini 2.0 Flash, GPT-4o-mini)</li> <li>Hosting: Vercel (frontend) + Railway (backend, $5/mo)</li> </ul> <p>Try it free: arborinquiries.com</p> <p>I'd love feedback — especially on the graph UX and research quality.</p> showdev ai react python Flynn L Tue, 24 Feb 2026 15:16:14 +0000 https://dev.to/flynn_l_e173fef9d85a8b4a5/show-dev-i-built-an-ai-research-dashboard-that-tracks-top-labs-daily-5a57 https://dev.to/flynn_l_e173fef9d85a8b4a5/show-dev-i-built-an-ai-research-dashboard-that-tracks-top-labs-daily-5a57 <h2> The problem </h2> <p>Keeping up with AI research across labs is impossible manually. arXiv is a firehose. Twitter is noise. Newsletters land 3 days late.</p> <p>I wanted one place to see what matters. It didn't exist.</p> <h2> What I built </h2> <p><strong>Lattice</strong> — an AI research intelligence dashboard that tracks what 20 top labs publish daily.</p> <p>🔗 <a href="https://www.layerthelatestinalattice.com" rel="noopener noreferrer">layerthelatestinalattice.com</a></p> <h3> Features </h3> <ul> <li>Daily paper feed from 20 tracked labs with AI summaries</li> <li>24 research topics with trend velocity charts</li> <li>Lab profiles with output sparklines and topic focus</li> <li>Weekly digest with most-read papers and breakout topics</li> <li>Collections with BibTeX/Markdown/JSON export</li> <li>No account needed — anonymous, localStorage only</li> </ul> <h2> One interesting technical challenge </h2> <p>Matching paper authors to research labs. Papers list affiliations as free text ("Dept. of Computer Science, University of X") — no standard format.</p> <p>The pipeline: OpenAlex API (by DOI) → arXiv HTML scraping → Semantic Scholar data → LLM extraction as a last resort. All matched against ~500 organizations. Hit rate varies wildly by source —<br> OpenAlex is great for established venues, arXiv scraping catches the rest.</p> <h2> Stack </h2> <ul> <li>Next.js 16 (Turbopack, standalone output)</li> <li>Supabase (Postgres + RLS)</li> <li>Vercel (hobby plan)</li> <li>GitHub Actions (daily 6am UTC cron)</li> <li>Semantic Scholar + arXiv + HuggingFace APIs</li> <li>OpenRouter (Gemini Flash) for summaries</li> </ul> <h2> Feedback welcome </h2> <p>What features would be most useful? What labs or topics am I missing?</p> <p>We're also live on Product Hunt today (<a href="https://www.producthunt.com/products/lattice-6" rel="noopener noreferrer">https://www.producthunt.com/products/lattice-6</a>).</p> ai machinelearning productivity showdev