Daksh Agarwal

ML engineer & researcher. MLE intern at Peakflo (YC W22), RA at NYU, published at ASIA CCS '26. LLM agents, RAG & vision. Open to collabs, roles & freelance

[Daksh Agarwal @Dakshagarwal76]

i'm daksh. i build ML systems, then try hard to break them with evals.

right now: MLE intern at Peakflo (YC W22) + research associate at NYU. LLM agents, RAG, computer vision.

here's what that's looked like so far 🧵

[Daksh Agarwal @Dakshagarwal76]

@OfirPress lmao the opposite of every race tip ever

[Daksh Agarwal @Dakshagarwal76]

@santosh79 agents that can only propose not execute is the right default imo. is the approval ux in the repo yet?

[Daksh Agarwal @Dakshagarwal76]

@kalyan_kpl love this. the missing piece in most rag papers tbh. is there a template people can just fill?

[Daksh Agarwal @Dakshagarwal76]

@prompt48 40/40 done with an empty cart is so cursed lol. are the fault injectors in the repo?

[Daksh Agarwal @Dakshagarwal76]

@vanshsoni yeah retrieval first or you're just tuning noise lol. do you eval them separately?

[Daksh Agarwal @Dakshagarwal76]

@EvanOtero hey, would love to apply. built azure ai search agents at grant thornton and do rag eval at nyu, pretty close to this. portfolio https://t.co/napPfaQN6N, can i dm?

[Daksh Agarwal @Dakshagarwal76]

@xeophon yeah if i can't reproduce it i don't really trust the score either lol

[Daksh Agarwal @Dakshagarwal76]

@HanyuShi43097 yeah scaffold is annoying but so worth it lol. do you have a preferred way to do it?

[Daksh Agarwal @Dakshagarwal76]

@Selen7005717917 sick, just jumped in. where should i start on the rfc?

[Daksh Agarwal @Dakshagarwal76]

@sujal0302 oh interesting, so it's more wrong expert hurts than right expert helps. what are you working on next for it? would love to help out if you need a hand

[Daksh Agarwal @Dakshagarwal76]

@cgironda the 70x claims fatigue is real lol. has anyone run it on actual long context tasks yet?

[Daksh Agarwal @Dakshagarwal76]

@The_Tradesman1 a heart emoji making claude write recursion is the funniest thing i've read today lol

[Daksh Agarwal @Dakshagarwal76]

@DibblaX es working on agent tasks without backprop is so cool. which tasks did you try it on?

[Daksh Agarwal @Dakshagarwal76]

@emollick next cycle every campaign probably has an agent writing its ads lol

[Daksh Agarwal @Dakshagarwal76]

@Lumen2088 honestly most memory papers never compare against a good rag baseline, glad someone did

[Daksh Agarwal @Dakshagarwal76]

@kig every dataset picking a different winner tracks with what i've seen too. are the intake cases public?

[Daksh Agarwal @Dakshagarwal76]

@FenghaiLi7 nice! what are you working on for it rn? would love to help out if you need an extra pair of hands

[Daksh Agarwal @Dakshagarwal76]

@rohanpaul_ai big banks blocking agents while the small ones build for them is kinda funny tbh