Hacker News: dippatel1994

New comment by dippatel1994 in "Ask HN: What is your (AI) dev tech stack / workflow?"

dippatel1994 — Fri, 05 Jun 2026 22:16:24 +0000

Thank you so much @mellosouls, means a lot!

New comment by dippatel1994 in "Ask HN: What is your (AI) dev tech stack / workflow?"

dippatel1994 — Fri, 05 Jun 2026 19:06:07 +0000

Thanks for spotting it but if I tried other way would have kicked out, it was a genuine try to help but thanks @mellosouls I can understand.

New comment by dippatel1994 in "Real-time LLM Inference on Standard GPUs: 3k tokens/s per request"

dippatel1994 — Fri, 05 Jun 2026 18:16:30 +0000

Exactly! Any optimization for local inference is a welcome change IMHO!

New comment by dippatel1994 in "Ask HN: What is your (AI) dev tech stack / workflow?"

dippatel1994 — Fri, 05 Jun 2026 18:14:05 +0000

Don't want to jeopardize this awesome chat about tools but for AI workshops I think these visual cards I came across could be an amazing way to handout. They cover all LLM concepts and explained visually. Found very useful to revise LLM concepts before AI research scientist/AI engineer interviews.

https://github.com/llmsresearch/llm-flashcards

Show HN: Chrome extension that tailors your resume to a job posting in one click

dippatel1994 — Fri, 06 Mar 2026 03:41:53 +0000

Job tailoring advice is everywhere but it all bottlenecks at the same place: actually doing it 50 times. Copy resume, paste into ChatGPT, rewrite bullets, reformat everything that broke, repeat. Ajusta sits on the job page (LinkedIn, Indeed, etc.), reads the posting, and rewrites your resume against the JD without touching your formatting or adding anything that isn't already there. The output is your resume with tighter language and better keyword alignment, not a new document built from a generic template. The constraint we cared about was hallucination. A lot of resume tools will invent responsibilities or rephrase something so aggressively it no longer reflects what you did. Ajusta only works with what's already in your resume. Still early. Would genuinely like feedback from anyone who tries it, especially on edge cases like non-standard formats or roles with heavy technical terminology.

Comments URL: https://news.ycombinator.com/item?id=47270574

Points: 1

# Comments: 0

Show HN: Open-source PaperBanana – academic diagrams from text via agents

dippatel1994 — Fri, 06 Feb 2026 13:35:14 +0000

The PaperBanana paper (arXiv:2601.23265) from Google Cloud AI Research and PKU describes a multi-agent framework for generating publication-ready academic illustrations from text. The official code hasn't been released yet, so I implemented it from the paper.

The pipeline chains 5 agents: a Retriever that selects reference diagrams, a Planner that generates a textual description, a Stylist that refines for visual aesthetics, a Visualizer that renders the image (Gemini for diagrams, Matplotlib for plots), and a Critic that evaluates and triggers iterative refinement.

Uses Google Gemini as the default backend (free tier works). Ships with an MCP server so you can use it directly from Claude Code or Cursor. Quick start: pip install -e ".[dev,google]" then paperbanana generate --input method.txt --output diagram.png

Repo: https://github.com/llmsresearch/paperbanana

This is an unofficial reimplementation and will differ from the original system. I plan to link to the official release once it drops. Happy to answer questions about the architecture or prompt engineering decisions.

Comments URL: https://news.ycombinator.com/item?id=46912649

Points: 1

# Comments: 0

New comment by dippatel1994 in "Show HN: Implementation of Google's PaperBanana (diagram generation from text)"

dippatel1994 — Thu, 05 Feb 2026 11:51:34 +0000

Released mcp server and skills support. You just need "uvx --from "paperbanana[mcp]" paperbanana-mcp" to configure paperbanana mcp server.

Show HN: Implementation of Google's PaperBanana (diagram generation from text)

dippatel1994 — Wed, 04 Feb 2026 16:16:14 +0000

The original authors haven't released code yet, so I built it from the paper. It takes a methodology section as input and generates a publication-style diagram.

The pipeline uses five agents: a retriever selects reference diagrams via in-context learning, a planner drafts the layout, a stylist adjusts for conference aesthetics, a visualizer renders with Gemini, and a critic evaluates and refines over three rounds.

The part that took the most effort was the reference dataset. The paper curates 292 (text, diagram, caption) tuples from 2,000 NeurIPS papers, filtering by aspect ratio and human review. Reproducing that required PDF layout extraction with MinerU, positional heuristics to identify methodology sections (paper headings are wildly inconsistent), and manual verification of each example.

Output quality depends heavily on reference set quality. Requesting community to submit their papers via issues so we can add them. Quality examples in, quality output out!

Runs on Gemini's free tier. Also includes an MCP server if you want to use it from your IDE. https://github.com/llmsresearch/paperbanana

Comments URL: https://news.ycombinator.com/item?id=46887669

Points: 3

# Comments: 1

New comment by dippatel1994 in "What ICLR 2026 Taught Us About Multi-Agent Failures"

dippatel1994 — Sat, 31 Jan 2026 15:29:33 +0000

Author here. I went through ICLR 2026 accepted papers looking for work relevant to multi-agent production problems. Found 14 papers clustered around 5 issues: latency (sequential API calls), token costs, error cascades, brittle topologies, and observability.

A few highlights: - Speculative Actions: parallel API execution, ~30% speedup - KVComm: share KV pairs instead of text, 30% of layers gets near-full performance - DoVer: intervention-driven debugging that flips 28% of failures to successes

Happy to discuss any of the papers or the framing. The decision matrix at the end maps each problem to a starting paper.

What ICLR 2026 Taught Us About Multi-Agent Failures

dippatel1994 — Sat, 31 Jan 2026 15:29:13 +0000

Article URL: https://llmsresearch.substack.com/p/what-iclr-2026-taught-us-about-multi

Comments URL: https://news.ycombinator.com/item?id=46837484

Points: 1

# Comments: 1

New comment by dippatel1994 in "Tell HN: Ask your AI, "What can you tell me that you know about me?" to see ..."

dippatel1994 — Tue, 27 Jan 2026 10:49:55 +0000

Add the word "Be Honest" and you will see the other side of the spectrum!

New comment by dippatel1994 in "Sam Altman said OpenAI planning to 'dramatically slow down' its pace of hiring"

dippatel1994 — Tue, 27 Jan 2026 10:46:55 +0000

Because they were hiring crazily!

New comment by dippatel1994 in "Agent framework selection became easy with this decision matrix diagram"

dippatel1994 — Mon, 26 Jan 2026 16:20:35 +0000

Last year, we evaluated different agentic LLM frameworks before selecting to architect our production system. Too many options out there.

So I built a covering their capabilities. If you're evaluating agentic solutions for your next project, hope this saves you time. It maps what's native, what needs integration, and what's not supported.

Let me know if anything needs correction or want to add any other framework.

Agent framework selection became easy with this decision matrix diagram

dippatel1994 — Mon, 26 Jan 2026 16:20:35 +0000

Article URL: https://media.licdn.com/dms/image/v2/D5622AQFGe8fWBYew_g/feedshare-shrink_2048_1536/B56Zv46WagIoAo-/0/1769407605435?e=1770854400&v=beta&t=r5LafTflp_8PhrdKjsW8kl37wRwgtl7WPtCfIu8wUN0

Comments URL: https://news.ycombinator.com/item?id=46767574

Points: 1

# Comments: 1