About
Specialties: Product insights, predictive modeling, analytics infrastructure, automated…
Activity
3K followers
-
Oliver Gothe shared thisOliver Gothe shared thisI’m hiring a PM on my team at Twitter to expand our efforts to bring transparency to political advertising around the world. https://lnkd.in/gcGTBnE
-
Oliver Gothe shared thisHi friends - I am hiring for a data engineer focused on revenue data at Twitter. I've really enjoyed my time here so far and we truly live by our motto #lovewhereyouwork. Let me know if you are interested or if you have any friends that might be looking for a new role. Join us! #dataengineering #data #twitter
-
Oliver Gothe shared thisHey friends - I am hiring data scientists focused on experimentation and optimization of Twitter’s ad marketplace algorithms. If you you have a passion for probability, modeling and solving complex and highly technical problems to drive value please let me know, I’d love to chat and tell you more about what we are up to and what 2019 has in store for us. #data #datascience #happynewyear
-
Oliver Gothe shared thisOliver Gothe shared thisGrocery stores prioritizing online ordering, home delivery as millennials take over market | The Star
-
Oliver Gothe shared thisI just wrote about Animation and Brand Design at Instacart and we're hiring designers!! https://lnkd.in/gv82Pgt
-
Oliver Gothe shared thisHow Instacart implemented HyperLogLog in Redshift and Tableau to calculate distinct counts 50 times fasterHow Instacart implemented HyperLogLog in Redshift and Tableau to calculate distinct counts 50 times faster
-
Oliver Gothe posted thisHey all - a good friend of mine is in the next YC batch and is working on an innovative Vegan startup. They're ex-Instacart/Yammer and are looking for a product-focused hacker to be the founding engineer. If you'd like to learn more, ping me and I'll pass along more info. #savetheplanet
-
Oliver Gothe shared thisCheck out my implementation of the HyperLogLog algorithm using just @Redshift and @Tableau. https://lnkd.in/gP3KDZaImplementing HyperLogLog in Redshift and Tableau – tech-at-instacartImplementing HyperLogLog in Redshift and Tableau – tech-at-instacart
-
Oliver Gothe reacted on thisOliver Gothe reacted on thisWe gave an AI 24 hours, 8 million tokens, and our legacy stack.✅ It felt like watching Star Trek. 🚀 Anthropic. OpenAI. Google... They dropped gravity on us. Most people are still finding their footing. Fair enough — this stuff is genuinely new. We started running. ⚡. Here's what nobody tells you — The context window isn't the problem. The problem is architecture. Most setups still treat AI as a conversation: prompt in, answer out, repeat. The session is the memory. And sessions decay. We stopped building that way 16 days ago. 𝗪𝗵𝗮𝘁 𝗰𝗵𝗮𝗻𝗴𝗲𝗱: 🧠 The spec became the memory — not the session 🎯 Agents take a mission, execute, report back, disappear 🔄 Context resets. Knowledge accumulates. 🛡️ No drift. No degradation. No hand-holding. Context limit during build-from-specs? Gone. Not managed. 𝗘𝗹𝗶𝗺𝗶𝗻𝗮𝘁𝗲𝗱. 📊 𝟭𝟲 𝗱𝗮𝘆𝘀 𝗼𝗳 𝗰𝗼𝗻𝘁𝗶𝗻𝘂𝗼𝘂𝘀 𝗼𝗽𝗲𝗿𝗮𝘁𝗶𝗼𝗻: ⏱️ 24h — single uninterrupted run 🔥 8M tokens — one session, zero drift 💎 67.4M tokens — 16 days I'm not going to pretend this was easy to figure out. It wasn't. But once it clicked — watching agents spin up, take their mission, execute in formation, report back — something shifted. Not in the tooling. In what's possible. We didn't optimise around context limits. We replaced the framework that created them. 🌍 The labs gave us the gravity. We just learned how to move inside it. If you're hitting the same walls — the answer isn't a bigger window. It's a different architecture. Happy to share what we learned. 💬 — Michal Bachorik #ClaudeCode #Anthropic #ChatGPT #Gemini #Google #ContextEngineering #SpecDrivenDevelopment #AutonomousAgents #AIEngineering #PersistentMemory #AgenticAI #LegacyModernization #BuildInPublic
-
Oliver Gothe liked thisOliver Gothe liked thisMeta Analytics team just published the playbook for building an internal AI analytics agent. In summary, context is the hard part. This largely echoes OpenAI's post in February. In six months, 77% of Meta data scientists were using the agent with 4,500+ community built recipes used 150K times. Feedback loops make it so that their users became collaborators. My take ways: - The headline stat that resonated with my experience is 88% of queries from Meta data scientists hit tables they already queried in the last 90 days. This allows for curation of context from millions of tables to dozens without sacrificing utility. - They built shared memory by running an offline LLM pipeline over every analyst's query history. The agent inherits colleague expertise by reading their past work with a powerful feedback loop. - Answers comes with the underlying SQL. Meta is building for data scientists who can read it. This enables interpretability for their target audience. - A separate AI validates outputs against domain rules before returning them. Trust is core! You cannot shortcut context. You need semantic models, workflows, governance, personal memory, validation, feedback loops and evals. 😉 Omni. Meta's playbook details what it takes. And how hard it is. Thanks Meta analytics team!
-
Oliver Gothe liked thisOliver Gothe liked thisLast Thursday was my last day at Meta, after almost 6 years working in FAIR and then GenAI. I leave with a huge lot of gratitude. Those papers we wrote, those large projects we built (Emu Video and Moviegen !) and those 6 years working with such talented and kind fellow researchers, in Montreal and all over the world, have been the highlight of my career so far. I’d like to thank Olivier Delalleau, Priya Goyal, Mannat Singh, Ishan Misra and Ce Liu for being such amazing managers and mentors to me over the last 6 years. I’d like to thank my former amazing close team, Andrew Brown, Xi Yin, Rohit Girdhar, Arun Mallya, Saketh Rambhatla, Samaneh Azadi, Yuval Kirstain, and all of you with whom I worked in Meta. You are all amazing. I’ll miss you. I am excited to share that I have joined AMI - Advanced Machine Intelligence, yesterday, as a Member of the Technical Staff, under Saining Xie and together with Michael Rabbat in Montreal. The team is super talented and very ambitious. I am super excited to open this new chapter and looking forward seeing the awesome things we are going to build together !
-
Oliver Gothe liked thisOliver Gothe liked thisIn San Diego for the NAGGL AI in 7(a) Lending Member Meetup, and it’s a full-circle moment. The San Diego Crew Classic race course start line is 500 feet from the conference venue. In a past life on the Cal rowing team, we’d come down here to test our early-season speed. And in many ways, it feels like a similar moment at OmniAI: a talented team, a lot of quiet work to build something that feels slick, and real excitement to see how it performs on the race course. Looking forward to catching up with other NAGGL members and digging into what AI can actually unlock in lending workflows.
-
Oliver Gothe reacted on thisOliver Gothe reacted on thisThis is what happens when you have an entrenched professional hiring culture that is completely divorced from aligning talents and skills with actual job requirements. This is the end product of a half a century plus of misguided policies.
-
Oliver Gothe liked thisOliver Gothe liked thisAnother AI startup just got gutted. And no one’s talking about who got left behind. Anthropic just acqui‑hired the co‑founders of Humanloop — Raza Habib, Peter Hayes, and Jordan Burgess—along with a core group of engineers. They did not acquire the product, company, or IP. Just the leadership and talent. This isn’t the first time. → September 2024 - Character.AI: Google rehired its CEO. The rest of the org didn’t follow. → June 12, 2025 - Scale AI: Meta poached founders. The original team got zero upside. → July 14, 2025 - Windsurf: Google licensed their tech and hired the founders. No equity. No control. No team. This is becoming Silicon Valley’s new normal: Recruit the founders. Abandon the people who built it with them. VCs call it “efficient.” Tech giants call it “strategic.” Founders call it “an exit.” But the teams? They call it betrayal. As a founder, I can't stop thinking about the quiet cost behind the headlines. What happens to the engineer who worked 80-hour weeks? What happens to the PM who helped land their first client? What happens to the designer who made the product feel like magic? They built the rocket, only to be cut out of the moon landing. If this is the culture we keep reinforcing, we’re not building a better tech ecosystem. We’re just reenacting the same patterns of gatekeeping and disposability. That’s why I’ve committed to building differently. At Jobright.ai, we’re building a product that respects talent—at every level. From interns to operators, everyone deserves to be seen, heard, and valued. Because the future of tech isn’t just AI. It’s people who still believe in building with integrity. If you’re reading this and you’ve ever felt disposable, I want you to know: There’s a different way. There’s still room for founders who protect their teams. And there’s still time to make Silicon Valley human again. Let’s stop clapping for exits that leave people behind. Let’s start celebrating the ones who bring everyone with them. 👉 Follow Jobright.ai — we’re building for builders.
-
Oliver Gothe liked thisOliver Gothe liked thisCome join an exciting, growing, mission-driven team and help us design safeguards for molten chloride fast reactors!
-
Oliver Gothe liked thisOliver Gothe liked thisWe’re hiring a Head of Data at Headway! This role is absolutely critical for where we are headed, let me tell you more. Since joining Headway earlier this year, I’ve been energized by the scale of the opportunity and the clarity of our mission. We’re building the technology and infrastructure to make mental health care more accessible, affordable, and effective for millions. Data is core to that work. We’re looking for someone to lead a 40+ person team across analytics, data science, and data engineering, reporting directly to me. But more than that, you’ll help turn data into one of Headway’s most strategic levers — powering better decisions internally and helping reshape how mental health care is delivered across the country. We’re looking for someone who’s worked at scale, understands the complexity of health data in fast-moving environments, and wants to build the systems and team that make data more reliable, scalable, and impactful. If you’re excited to build for both urgency and long-term impact, I’d love to connect (or hear from someone you’d recommend). https://lnkd.in/eFg6AM6m #hiring #data #healthtech #mentalhealth
Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Education
-
University of California, Berkeley
3.9
-
Activities and Societies: San Francisco Python Meetup
Key Accomplishments
- Characterized the production and decay properties of the heaviest elements using the cyclotron at Lawrence Berkeley National Laboratory.
- Discovered a new vibrational level in fermium-254 in the nuclear reaction 48Ca+208Pb → 254No.
- Programmed and maintained large C/C++ programs to process and analyze extensive data sets.
- Developed effective oral and written communication skills through internal and external research presentations, as well as writing…Key Accomplishments
- Characterized the production and decay properties of the heaviest elements using the cyclotron at Lawrence Berkeley National Laboratory.
- Discovered a new vibrational level in fermium-254 in the nuclear reaction 48Ca+208Pb → 254No.
- Programmed and maintained large C/C++ programs to process and analyze extensive data sets.
- Developed effective oral and written communication skills through internal and external research presentations, as well as writing of proposals, standard operating procedures, and research articles.
- Organized and led diverse international research collaborations for experiments at the accelerator.
- Supervised, mentored, and trained a research associate and several undergraduate and graduate students.
- Taught three undergraduate courses for the chemistry department at UC Berkeley.
-
-
-
Licenses & Certifications
-
-
Practical Machine Learning
Coursera Course Certificates
Publications
-
Nuclear spectroscopy of the heaviest elements: studies of 254-No, 257-Rf, and 261-Sg
J. Phys.: Conf. Ser. 312
-
New Superheavy Element Isotopes: 242-Pu(48-Ca,5n)285-114
Physical Review Letters 105(18)
Patents
-
Populating Catalog Data with Item Properties Based on Segmentation and Classification Models
US 16/048,800
Other inventors -
Courses
-
Inferential Statistics
-
Honors & Awards
-
Barry M Goldwater Scholarship
Barry Goldwater Scholarship and Excellence in Education Program
The Barry Goldwater Scholarship and Excellence in Education Program was established by Congress in 1986 to honor Senator Barry Goldwater, who served his country for 56 years as a soldier and statesman, including 30 years of service in the U.S. Senate.
Languages
-
English
Native or bilingual proficiency
-
German
Native or bilingual proficiency
Other similar profiles
Explore more posts
-
Rohit Shekhar
There is Bad AI. And then there is Chiron. Not all AI is created equal. Bad AI: Query time reasoning on raw telemetry. Latency compounds at scale. Black box inference with no causal trace engineers can inspect. Alert volume stays high. Correlation happens after the fact. Infrastructure costs scale linearly with data volume with no ceiling. Then there is Chiron: Causal rules compiled at build time. This delivers sub 500ms deterministic RCA for the fastest possible MTTR. Single actionable alert per incident. Causality is determined before impact. Every alert carries a full evidence chain back to the root cause. Query store free architecture. A streaming state store runs alongside existing infra with no rip and replace. Compressed state over raw logs. The cost profile is decoupled from data volume. SREs have been trained to react faster. The better engineering problem is asking why we are reacting at all. That is the architectural shift Chiron is built on.
2 Comments -
Maria Vechtomova
Databricks is decommissioning online tables. They’ll be gone after January 15, 2026. Online tables have been the go-to for feature serving, including scenarios when your model serving endpoint does feature lookups at inference time (probably the most common pattern for production ML). The suggested path forward is to migrate to the new Databricks Online Feature Store. But it’s not exactly seamless: ➡️ Updating online tables via the SDK can fail unpredictably ➡️ You may run into strange data type conversion errors (int32-int64 conversion) It makes me wonder, who’s actually using online tables right now (they never left Public Preview), and how painful will this migration be for them?
15 Comments -
Andrei Lopatenko
Document-centric tasks sit at the core of many enterprise, business, and government workflows. Search is important, but the real challenge is going beyond retrieval, enabling systems to reason across documents, verify facts, and handle multi-step information tasks. Great to see a new model from Databricks moving in this direction. I expect we’ll see many more models, including open-weight ones, designed specifically for document-driven workflows, an area with huge potential across enterprise and government use cases.
1 Comment -
Galen Marchetti
AstroBee generates ontologies...but what is an ontology, and if they're important, why don't you already have one? I've written a short piece on Kimball star schemas, ontologies, and how each relate to Bill Inmon's original approach to data warehousing in the early 90s. Dig in to find an overview of a fascinating technological debate and the re-emergence of an approach to data modeling that's been sleeping for 20 years.
3 Comments -
Chris Huff
Snowflake’s SnowWork positioning is a signal. The race isn’t just about data anymore...... but about owning the AI workflow layer. Because once you control the workflow, you control how decisions get made. But here’s the gap: Workflows amplify whatever you feed them. And in most enterprises, the weak point isn’t orchestration. It’s the inputs..... messy documents, missing context, no clear lineage. So you get faster automation, without defensibility. That’s the real tradeoff showing up at the exec level: speed vs. trust The way I’m thinking about it is in three layers: 1. Workflow (orchestration) 2. Models (reasoning) 3. Inputs (can I trust what goes in?) Because you don’t get reliable AI by improving workflows. You get it by controlling what enters them. As platforms push toward owning the AI control plane, where are you placing your bets? If you're exploring this idea of the “AI control plane,” here is a simple breakdown: https://lnkd.in/gc2_4r97? #EnterpriseAI #DataStrategy #AILeadership #Snowflake #DigitalTransformation #AIControlPlane
4 Comments -
Pete Soderling
General-purpose LLMs can write SQL. But when Snowflake tested them against messy production data - malformed timestamps, evolving schemas, dialect-specific constructs - they topped out around 40-44% execution accuracy. So Snowflake trained their own. Arctic-Text2SQL-R1.5 is a specialized reasoning model purpose-built for Snowflake SQL. It outperforms GPT-5, Claude Sonnet 4.5, and Gemini 2.5 Flash on their internal benchmarks at up to 3x lower latency. This is the build-vs-buy decision in action as a shipped product powering Snowflake Intelligence in production. And Gaurav Nuti helped build it. At AI Council 2026, he's sharing the framework Snowflake uses to decide when to train their own models and what they learned the hard way doing it. "How to Unlock Enterprise Value by Training Your Own Language Models" May 12–14 | San Francisco → https://lnkd.in/eBCieAWH
2 Comments -
Sam Mahdavian
Everyone’s talking about frontier LLMs like GPT‑5, Claude Opus 4.5, and Gemini 3. Most people are missing the point! The real competitive edge isn’t access to a new model. It’s the ability to build the entire production system around it. It’s the engineering that turns a flashy demo into a trusted, efficient, scalable workflow: • 𝗗𝗮𝘁𝗮 & 𝗢𝗿𝗰𝗵𝗲𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻: High‑quality data pipelines feeding clear prompts, tools, and policies • 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻 & 𝗠𝗼𝗻𝗶𝘁𝗼𝗿𝗶𝗻𝗴: Rigorous testing, continuous monitoring for drift, and human‑in‑the‑loop review • 𝗚𝘂𝗮𝗿𝗱𝗿𝗮𝗶𝗹𝘀 & 𝗙𝗶𝗻𝗢𝗽𝘀: Strong security and compliance, paired with relentless cost optimization The future belongs to a few key groups working together: • 𝗧𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝘀𝗽𝗲𝗰𝗶𝗮𝗹𝗶𝘀𝘁𝘀 who can design secure, reliable, and efficient LLM (and agentic) pipelines • 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝘀𝗽𝗲𝗰𝗶𝗮𝗹𝗶𝘀𝘁𝘀 who deeply understand their domain and can “𝘀𝗽𝗲𝗮𝗸 𝗟𝗟𝗠” • 𝗣𝗿𝗼𝗱𝘂𝗰𝘁/𝗼𝗽𝗲𝗿𝗮𝘁𝗶𝗼𝗻𝘀 𝘀𝗽𝗲𝗰𝗶𝗮𝗹𝗶𝘀𝘁𝘀 who translate messy real‑world needs into clear requirements, guardrails, and success metrics Individually they’re powerful, but it’s the hybrid builders and cross‑functional teams at the intersection of these three that turn frontier models into durable, production AI. They don’t just ship another chatbot. They ship real outcomes. What’s the most overlooked component in building production AI? #AI #GenerativeAI #LLM #AILeadership #AIEngineering #AArete
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content