unfair.so
Browse more creators
@pankajkumar_dev
Pankaj Kumar
Pankaj Kumar16.2K
Pankaj Kumar
This creator added themselves. Not verified by unfair.so.
@pankajkumar_devAI ModelsAI AgentsAI Development

I build things | Dm for work/collab

16.2KFollowers
20.4KMedian impr. / post
0.9%Engagement
8+3
Paid partnerships
In their audience
3+ from @ycombinator follow Pankaj Kumar
@NickK2305@ojhaverse23@vrn21_3+ affiliated with @superteamin2+ affiliated with @googledeepmind2+ from @mit2+ affiliated with @openai2+ affiliated with @aoagents
1,500+ verified followers analyzed
@NickK2305@ojhaverse23@vrn21_+1497107+ AI founders107+ AI engineers80+ engineers80+ AI creators49+ founders35+ AI researchers
25–34 core age group
36.3%
15.4% of audience in United States
🇺🇸
Last 7 days66 posts analyzed · replies not counted · updates weekly
ImpressionsEngagements
272K136K0
2.4K1.2K0
Aug 30Aug 31Sep 1Sep 2Sep 3Sep 4Sep 5
Past collabs
Pankaj Kumar@pankajkumar_dev · Aug 14Gemini 3.7 Flash at 50% off on OpenRouter Now it actually makes sense to use it for running agents at scale. Multimodal + agentic capabilities without having to worry as much about the bill.GooglePaid partnership7.2K views
Pankaj Kumar@pankajkumar_dev · Aug 12I asked ChatGPT for a cinematic shot. Higgsfield Plugin is live in the ChatGPT app, so the finished result came straight back into the conversation. The whole generation happens right there. Cost: one message per iteration. Worth it.HiggsfieldPaid partnership18.1K views
Pankaj Kumar@pankajkumar_dev · Aug 4I wasn't expecting the most interesting part to be the writing. It was the fact that the AI grades its own output before publishing. That's a workflow I don't see people sharing often.BlotatoPaid partnership6K views
Pankaj Kumar@pankajkumar_dev · Jul 30I asked MiniMax M3 using Claude Code via API to build a green city of the future. Test + review: works really well with Claude Code: one-shot builds, testing, and fixing across a full project. Strong coding + agentic flow, long-context handling for the entire site in one session, and image generation is good too (using the MiniMax API via a script, since Claude Code doesn't support image generation). All of this sits on MiniMax Token Plan Max ($50/mo) M3 + image + speech + music share one quota / one key (5.1B M3 tokens/month). 15 sections: hero, manifesto, living nature, 6 districts, daily life, architecture, energy, transit, wildlife, labs, resident stories, city systems, a day-in-the-city. Every image generated by MiniMax too: 37 cinematic renders across the whole site. Worth it.MinimaxPaid partnership7K views
Pankaj Kumar@pankajkumar_dev · Jul 21Intelligence keeps getting cheaper every week, but this is the first time I've seen the same model just cost less. Ship by @thesean_labs: change model="claude-opus" to model="ship-like/claude-opus" and pay a fixed 50% less. Same quality, half the bill. http://app.thesean.aiWithmartianPaid partnership16.2K views
Pankaj Kumar@pankajkumar_dev · Jul 19Kimi K3 vs Claude Opus 4.8 $0.23 vs $0.25 Same prompt: build a launch site for an architecture studio. Kimi: blue-hour glass house, warm light pouring out, someone inside looking out, reflective pool, blueprints drawing it up. Finished. Lit. Opus: beautiful type. A translucent tan volume. Wood lattice. No light, no life, no place. Not a finished building. Same cost. Completely different building.KimiPaid partnership58.7K views
Pankaj Kumar@pankajkumar_dev · Jun 10I got tired of "is this startup idea good?" taking 3 hours of random Googling. So I built a 6-step startup validation agent powered by Step 3.7 Flash @StepFun_ai plan → competitor research → product mapping → saturation score → SWOT → differentiation One input: AI Resume Roaster One output: verdict + competitive wedge + 30-day execution plan What stood out wasn't just the final output. In my test, Step 3.7 Flash completed the workflow in around 1-2 minutes, searching 7-8 sources across 6 reasoning steps from research and analysis to positioning and strategy without me bouncing between multiple tools. The search capabilities were especially useful because research became part of the reasoning process instead of a separate step. For workflows like this, what matters isn't a benchmark score its whether the model can actually complete the task end to end. It saved me from jumping between multiple tabs, tools, and prompts. Try it: https://startup-idea-validator-two.vercel.app/ Step 3.7 Flash: http://platform.stepfun.aiStepfunPaid partnership5.7K views
Pankaj Kumar@pankajkumar_dev · May 18Voice AI really went from multi-month enterprise rollouts to "lemme try this real quick"PolyPaid partnership6.7K views
Planning a launch?Pankaj Kumar is one of 200 curated B2B creators. Tell us what you are launching and we will put the shortlist together.
[email protected]·t.me/fdrtrm
Built with unfair.so
1 picked$400