TRXND
live content distribution engines i'm building. main site, app pipeline, reddit intelligence.
signal intelligence and live context for ai agents, not another static data feed.
one thing that becomes painful quickly is signal overload.
once retrieval gets easy, customers drown in hiring signals, company updates, people movement, enrichment events, trigger streams.
the hard question stops being how do we get data.
it becomes: what actually matters right now?
a ranking layer on top of crustdata apis that scores buying intent, urgency, behavioral anomalies, company momentum, and signal decay.
cluster related events, generate momentum scores, summarize why something matters, recommend next actions. actionable sequencing, not raw dashboards.
example: funding + infra hiring + ai feature launch + traffic spike + new vp engineering should rank differently than isolated events.
- sales automation + onboarding copilots
- customer research agents
- auto-generated demos
- technical support copilots
- gtm infrastructure + workflow generators
i like building systems that remove operational pain.
usually i see something inefficient, obsess over it for a few hours, and ship a fix before people finish discussing the problem.
ai-native generalist. i see friction, build systems around it, ship fast, and work across product, infra, and gtm when things are undefined.
- live content distribution engines at trxnd · shipping
- founder's office at saturn labs · bangalore
- founder's office at aevis · remote
- overall convener, inter iit tech meet 14.0
crustdata feels early in the right way.
the internet was designed for humans clicking links.
now agents need live context, structured retrieval, real-time signals, and systems that can actually act on information.
that shift is massive.
most companies are still thinking about getting data.
the harder problem is figuring out what actually matters once the data starts flooding in.
that's the kind of problem i like working on.
at saturn labs, hundreds of hours of robotics training data were uploaded daily.
the problem wasn't collection. it was qc.
people were manually checking:
- hand visibility
- lighting quality
- object overlap
- frame usability
- motion consistency
repetitive work. huge time sink. draining engineering bandwidth every day.
so i built an automated qc system overnight.
it scored videos, flagged low-quality data, and integrated directly into the pipeline the same night.
next day the team stopped wasting hours on manual checks and focused on actual engineering work instead.
if something repeatedly slows people down, i automate it.
- saturn labs + founder's office: priorities changed daily, ownership fluid
- inter iit tech meet 14.0: 23 iits, 2000+ participants, ₹1cr+ ops
- not waiting for scoped tickets. ambiguity and speed are normal.
things i've built
IronClaw
multi-agent android automation using accessibility trees, not apis.
SpeakLingo
real-time voice cloning translator, sub-545ms latency on live calls.
Mini-vLLM
inference engine from scratch, because i wanted to understand serving.
CiteAgent
agentic rag for causal extraction. 5th at inter-iit tech meet 14.0.
SoulScript
ai wellness platform with real-time avatars and memory systems.
i don't like unnecessary complexity.
i like systems that:
- reduce friction
- remove repetitive work
- increase speed
- help people focus on meaningful problems
i think speed is underrated.
i think most meetings should've been documents.
good operators are usually hidden behind the scenes quietly removing chaos.