Kellen Donohue

If you are in the Bay Area and run Growth or GTM you won't want to miss this 👀

3

1 Comment

I’ve seen how traditional data sharing can slow teams down through heavy provisioning, delayed analytics, and long deal cycles. This past year, I had the opportunity to work with Jaylyn Stoesz, Ibotta and our Databricks partners to level up how we approach secure data collaboration. This post shares my perspective of how modern solutions such as clean rooms are changing the data collaboration landscape.

39

1 Comment

Our team won an internal AI Hackathon at CoBank last week. Here's what we built. The problem: Extracting insights from our loan portfolio data requires SQL expertise most business stakeholders don't have, and shouldn't need. What we built: A natural language reporting tool on Amazon Bedrock. You ask a plain-English question, and it generates validated SQL, runs it against our data lake, and produces a formatted executive report. No SQL required. The key insight: any AI model is only as good as the data you give it. A model can read a schema, but it doesn't know which tables can be joined, on what keys, or what the results actually mean in the business context. We embedded our domain knowledge (the data model, table relationships, and business logic) directly into the knowledge base so the agent could reason across our data model the way an experienced analyst would, to extract real insights. Stack: Amazon Bedrock (Agents + Claude 3.5 Sonnet), Athena, Glue, Streamlit, S3 Credit to the team that made it happen: Joshua Todd, Quyen Tsai, Pravin Shetty, Utham K. Strong work all around.

35

4 Comments

I've gotten so many notifications about my credentials are now out in the market for anyone to pick up. Am I worried? Not exactly because I have MFA to help mitigate someone impersonating me or holding my accounts as ransom. Your enterprise apps shouldn't be any different, if anything your InfoSec team are working hard to ward of cybersecurity attacks from all angles. MFA helps with that. Check out the latest update from Snowflake on their MFA roll out. #SnowflakeSquad #SnowflakeSquadMember #SquadMember #SquadGoals

3

I've been sharing my thoughts on agents lately. I've written an expanded article with more direct thoughts, processes, and how I think about what I've settled on calling Rails-Driven Development. How does this technique compare to how you use agents? https://lnkd.in/eNGPrggG

4

Our VP of Data Analytics, Phil, wrote a very accessible blog post about another in-house agentic solution built at Fullscript. Flurry allows anyone on the team to use natural language to ask a data question, and get back accurate, usable, and understandable information. This gives our data team more time to work on the interesting stuff! https://lnkd.in/ekmdfbJg

3

Practical architectural choices are still the job. I’m enjoying LLMs and coding agents as much as anyone - they unquestionably help us move faster. They plan, scaffold, refactor, and can set off and produce with whatever bounds we give them for the most part. But production systems still come down to judgment. Practicality. Restraint. And knowing what you’re willing to evolve and maintain. I’m usually building more than I write about, but I recently built a small markdown publishing engine inside Rails. Not a massive architectural decision, but decoupling cache life-cycle from deploy life-cycle felt worth sharing. Engineers make dozens of decisions like this on every project. Some make life easier. Some quietly make it harder. The interesting part isn’t Redis or Solid Cache. It’s knowing when to evolve, and when not to over-engineer, based on the goals of the system or project. Personally, I’m enjoying this era: focusing more on architectural judgment, while letting agents handle more of the mechanical pieces. Here’s the write-up: https://lnkd.in/eTDsGNZM

11

Imagine knowing which customers might churn. Imagine knowing which customers are ready to grow. Imagine knowing this before they know. 🤯 That’s the power of Forwrd.ai, and I’m so excited that through our latest acquisition, we’re bringing this capability to the Pendo.io platform. Pendo Predict is coming, and it’s going to help teams turn usage data into real business outcomes. Read the full announcement from our CEO, Todd Olson.

1

🧠 AI Rights Historical Wrongs Researchers at Stanford and Princeton have fine-tuned a large language model to uncover racially discriminatory housing clauses buried in millions of historical property deeds — accelerating a process that could have taken a decade by hand. 📍 The problem: Many property deeds in Northern California — especially in Santa Clara County — still contain illegal racial covenants from as far back as the 1850s, banning people of color from owning or living in certain homes. 🔎 The solution: The team used OCR to digitize 5.2 million deed pages, then fine-tuned Mistral-7B via LoRA on a hand-labeled dataset of discriminatory phrases. The model learned to identify harmful language in context, avoiding false positives like the surnames "Black" or "White." 📊 Results: Detected 24,500 lots with racial clauses (~25% of homes in 1950) Found that 10 developers were responsible for 1/3 of all discriminatory clauses Model reviewed all pages in 6 days at a cost of $258 Equivalent manual effort: 10 years, $1.4M 💥 This work doesn't just use LLMs to build the future — it uses them to understand and repair the past. The open-sourced model now enables other U.S. counties to follow suit. #AI #MachineLearning #LLM #ResponsibleAI #Stanford #Princeton #HistoryMeetsAI #TechForGood #EthicalAI #SocialImpact #PropertyTech https://lnkd.in/dtec-zbx

4