Aditya Parameswaran
@adityagp
Followers
5K
Following
3K
Media
104
Statuses
2K
associate prof @ucberkeley, co-director @ucbepic, cofounder @ponderdata (acq. @snowflakeDB) | on a mission to make data science effortless at scale | he/him
Berkeley, CA
Joined April 2008
SIGMOD updates: slides from my NOVAS workshop keynote on the next frontier of 🌟LLM-powered🌟 data tooling — including the document "stack stack"^TM that we've been cooking up @UCBEPIC , plus. Slides from the panel on the future of LLMs-meets-DB, including my predictions on how
1
1
4
.@united is it your official policy to discriminate against parents of young children? Even though we boarded early in UA 59, (about to take off) in group 1 (thanks to status) and had found space overhead for our stroller, one of your staff came later and forced us to check it in.
1
0
5
A decade of HILDA! . Still a vibrant sub community of databases, human in the loop is more relevant than ever. From Arnab’s keynote.
0
0
6
RT @mdancho84: 🚨BREAKING: New Python library for agentic data processing and ETL with AI. Introducing DocETL. Here's what you need to kno….
0
183
0
RT @alvinkcheung: In this work, we show how to use off-the-shelf LLMs to generate code for accelerators. This is interesting as accelerator….
0
3
0
RT @LuisvonAhn: I came to the US on an F-1 student visa. Since then, I’ve built a company worth over $20 billion—headquartered in Pennsylva….
0
600
0
Excited to partner with the PromptQL folks on this important benchmarking effort around data agents!.
.@PromptQL partners with @UCBerkeley to develop a new Data Agent Benchmark for Reliability of Enterprise AI Agents. Current benchmarks test lab scenarios, not real business complexity. They miss siloed data, compliance needs, and mission-critical pressure. We will build in.
0
1
4
RT @IsilDillig: 1/3 The US didn’t end up leading the world in computing by luck. It happened because it made long-term, public investments….
0
38
0
Low-cost LLM-powered data processing at scale **with guarantees** - now open-source! . Get accuracies that are nearly that of your SOTA model at a fraction of the đź’¸. Take a look and let us know what you think.
🚨 📢 Releasing BARGAIN: Guaranteed Accurate AI for Less. 💰 BARGAIN reduces LLM powered data processing costs by using cheaper LLMs (e.g,gpt-4o-mini) on data records they are accurate on .🎯 It *theoretically guarantees* the output matches expensive LLM's output (e.g,gpt-4o)
0
1
5
Shreya @sh_reya is one of the world's leading experts on LLM evals. she's able to speak fluently to both academics and practitioners alike - and has led a bunch of foundational research in this area (spade, evalgen, promptevals). Check this out!.
we’re teaching a course for engineers and technical PMs on evals for AI products! why?. despite all the tooling out there, there’s very little *education* on how to build evals—like, what’s the actual process from vague idea to deployed product? how do you know if your system is.
0
4
30
We've been working on extracting information in templatized PDFs for the last couple of years, leveraging the best of LLMs and classical data extraction techniques. Our latest technique, TWIX, has the best of all worlds: beats Azure DI, AWS Textract, or LLM-based approaches by.
28
88
844
RT @WHinthorn: Constrained generation undergirds most LLM workflows, but most folks don't take the time properly ensure instructions are fo….
0
3
0