The NeuralPod

The NeuralPod

By Chris CoyneScienceTechnology
Download on the App Store

All The NeuralPod episodes

  • The One Person Go-to-Market Army for B2B Founders, Tim Scheuer on Driving Sales with AI Agents

    How can B2B founders use AI agents for sales, lead generation and outbound? Tim Scheuer, cofounder of OXYGEN, explains how he runs outbound, content, SEO and ads as a one person go to market team.

    In this episode of The NeuralPod, Christopher talks to Tim about go to market (GTM) engineering: making sales and marketing run more like software, so agents can do useful commercial work from company context and buying signals. Tim explains why needing nine different tools led him to build OXYGEN, an agentic revenue platform that brings lead and signal sourcing, multichannel outbound, CRM and agents into one place.

    He also outlines his playbook for a company that has just raised its Series A, and is clear that results take testing: he describes first positive conversations within two to four weeks, with stronger results after two to three months of refining segments and messaging.

    IN THIS EPISODE
    • Why Tim sees go to market becoming more like software, and why that is driving demand for technical GTM engineers.
    • How an "allbound" motion combines inbound and outbound, and how the signals it creates feed the next round of outreach.
    • The warm signals he prioritises before cold outreach, from qualified signups to post engagement and profile views.
    • Why his "second brain" is built for his agents, and how company context stops them going wrong.
    • How far he trusts agents with outbound and his inbox, including an agent that fixed a bug and replied without him.
    • Why he believes agents trained with reinforcement learning will keep pushing a metric such as reply rate when run in a loop.

    "The second brain is not for me. The second brain is for my agents."

    CHAPTERS
    00:00 Cold open: a one person go to market army
    00:31 Meet Tim Scheuer, cofounder of OXYGEN
    01:49 AI agents and the rise of GTM engineering
    04:49 Running sales and marketing as a one person team
    06:18 The Series A playbook: inbound, outbound and signals
    08:57 How long does AI outbound take to work?
    10:34 Building a second brain for AI agents
    12:07 Trusting agents with outbound and your inbox
    13:41 Outbound volume and the buying signals that matter
    16:49 Customer results and onboarding founders new to AI
    20:36 Agent routines, loops and reward hacking
    23:02 OXYGEN's goals and recommended books
    24:50 Closing thoughts

    GUEST AND RESOURCES
    OXYGEN: https://oxygen-agent.com/
    Books mentioned:
    • Founding Sales by Pete Kazanjy
    • Scientific Advertising by Claude C. Hopkins

    FOLLOW THE NEURALPOD
    Subscribe for conversations with the founders, researchers and engineers building and applying AI.
    YouTube: https://www.youtube.com/@TheNeuralPod
    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-neuralpod/id1761461659
    The NeuralPod is hosted by Christopher Coyne, founder of NeuralRec: https://neuralrec.ai

    Recorded 23 October 2026.

    #AI #AIAgents #AISales #GTMEngineering #AgenticAI #B2BSales #OutboundSales #SalesAutomation #BuyingSignals #FounderLedSales #GoToMarket #LinkedInOutreach #StartupSales #SecondBrain #Allbound

    26 min
  • What It Takes to Build a British LLM and How the UK Gets Ahead in Sovereign AI with Adie Holt

    What does sovereign AI really mean for the UK? Adie Holt of Cosine explains why the company is building Lumen Sovereign, a British large language model trained on the Isambard AI supercomputer in Bristol, and what the country needs to stay in the AI race.

    On The NeuralPod, Christopher and Adie cover his 24 year engineering career in the Royal Air Force, the leadership idea of confident humility, and Cosine's move from a not for profit research lab to code engineering for regulated industries such as banking and defence. Adie defines sovereignty as freedom of action: guaranteed access to, and control over, the AI models and systems you depend on.

    IN THIS EPISODE
    • Why Adie sees relying solely on trusted access to third party models as riskier than ever, without asking anyone to switch off existing tools.
    • How he argues the UK can compete without the biggest labs' budgets, now that training methods have become more efficient.
    • Isambard's footprint as Adie describes it: 5 megawatts rather than 50, and one swimming pool of water a year.
    • Sparkler, Cosine's training tool, which Adie says has produced a 200 billion parameter version of Lumen Sovereign and is training a trillion parameter model.
    • Data provenance, independent safety testing, and coalition partners who, he says, have shared data so the model handles UK tax law and speaks Welsh.
    • What Adie wants next: distributed compute built with local communities, and government backing its winners.

    Adie Holt: "If we back artificial intelligence today, we put ourselves on the top table of one of the most important and defining technologies of the next 100 years."

    CHAPTERS
    00:00 Cold open: Adie's prediction for UK AI
    01:06 Welcome, and from the RAF to defence innovation
    03:49 How the military and corporates adopt technology
    06:37 Leadership lessons: confident humility
    09:22 What sovereign AI means to Cosine
    13:09 Why trusted access may no longer be enough
    14:59 Can the UK compete on cost and compute?
    19:12 Cosine's story: code engineering for regulated industry
    21:41 Trust, data provenance and safety testing
    23:57 Sparkler and training on Isambard
    26:09 Lumen Sovereign, its coalition and agency
    29:28 Predictions: can Britain stay in the AI race?
    32:19 What the UK should do next
    34:39 Book recommendations and closing thoughts

    GUEST AND RESOURCES
    Adie Holt, Defence Lead at Cosine at the time of recording
    Cosine: https://cosine.sh
    Books mentioned:
    • So Good They Can't Ignore You by Cal Newport
    • Incorruptible by Eric Ries

    FOLLOW THE NEURALPOD
    Subscribe for conversations with the founders, researchers and engineers building and applying AI.
    YouTube: https://www.youtube.com/@TheNeuralPod
    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-neuralpod/id1761461659
    The NeuralPod is hosted by Christopher Coyne, founder of NeuralRec: https://neuralrec.ai

    Recorded 30 September 2026.

    #SovereignAI #AI #UKAI #LargeLanguageModels #LumenSovereign #IsambardAI #Cosine #BritishAI #AICompute #DefenceTech #AISafety #DataProvenance #AirGappedAI #RegulatedIndustries #AIPolicy

    39 min
  • Achieve AI Agent Success in Enterprise from Pilot to Production with Darminder Singh PhD

    How do you take agentic AI in the enterprise from pilot to production, and where does it deliver a measurable return? Darminder Singh, Director of AI at HCLTech, explains where AI agents create value and why process, culture and people often matter more than the technology.

    In this episode of The NeuralPod, Christopher talks with Darminder, who holds a doctorate from UCL, studied at London Business School and has worked in data science and AI at KPMG and Fujitsu, including defence. At the time of recording he had been at HCLTech for around ten months, where its AI labs act as a technology agnostic advisor to clients.

    IN THIS EPISODE
    • Where agents deliver a return on investment (ROI): start from a process bottleneck, understand the end to end workflow and get orchestration right.
    • Why repeatable, measurable results build the trust agents need to work by themselves.
    • Lessons from defence, from isolated systems at the edge to change management in the back office.
    • Common mistakes, from poorly defined problems and conflicting key performance indicators (KPIs) to siloed data, and how the Model Context Protocol (MCP) supports a data mesh approach.
    • Why agent memory gives vector databases and knowledge graphs a new purpose.
    • Guardrails, sandbox red teaming and risk based scoring for agent security.

    Darminder also shares what he looks for when hiring, why the context you give an agent may become your key differentiator, and why he sees work moving towards autonomous AI employees, without putting a timeframe on it. He closes on open source models, citing recent results where Google's Gemma outperforms Qwen on certain task based benchmarks.

    As Darminder puts it: "I think we've gone past a phase where a lot of our clients are doing pilots and experimentation."

    CHAPTERS
    00:00 Cold open
    00:23 Welcome and Darminder's career so far
    02:26 Research, business school and consulting lessons
    05:36 Advice for breaking into AI teams
    08:03 Deploying AI in defence
    11:48 Cloud ecosystems, Copilot and software engineering
    15:11 HCLTech's AI labs and agents in production
    17:44 Where agents deliver ROI
    21:42 Agent memory, vector databases and knowledge graphs
    23:30 Common mistakes: KPIs, siloed data and MCP
    26:23 Agent security, guardrails and red teaming
    28:19 Agent skills, context and Darminder's own tools
    31:37 How close are autonomous AI employees?
    32:45 Open source models and closing thoughts

    GUEST AND RESOURCES
    Darminder Singh, HCLTech
    Mentioned: Microsoft Copilot, OpenClaw, Claude Code, OpenAI Codex, MCP, Google's Gemma and Alibaba's Qwen.

    FOLLOW THE NEURALPOD
    Subscribe for conversations with the founders, researchers and engineers building and applying AI.
    YouTube: https://www.youtube.com/@TheNeuralPod
    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-neuralpod/id1761461659
    The NeuralPod is hosted by Christopher Coyne, founder of NeuralRec: https://neuralrec.ai

    Recorded 10 April 2026.

    #AgenticAI #EnterpriseAI #AIAgents #AI #AITransformation #AIStrategy #AIROI #AIGovernance #ChangeManagement #KnowledgeGraphs #MCP #DefenceAI #OpenSourceAI #AILeadership #HCLTech

    35 min
  • Are Your AI Agents Observable? Monitoring OpenClaw With ClawMetry with Vivek Chand, Founder

    Why do AI agents like OpenClaw need observability, and how can you see what yours is actually doing? Vivek Chand explains how he built ClawMetry, an open source observability dashboard for OpenClaw, and why visibility matters once agents can act on your computer.

    In this episode of The NeuralPod, Christopher talks with Vivek, who at the time of recording was a tech lead at Booking.com, leading customer support AI work after years in startups. ClawMetry sits next to OpenClaw and reads the files it logs to, showing what it is thinking, which files it accesses, which subagents it spawns and how many tokens it uses. Christopher notes it had reached 28,000 downloads at the time of recording.

    IN THIS EPISODE
    • Why Vivek sees evaluation loops as what turns an easy AI prototype into a real return on investment.
    • Why agents acting in parallel need different observability from services monitored with Grafana and service level objectives (SLOs).
    • How OpenClaw's long term context changed his coding, and his estimate that ClawMetry took a weekend to three or four days with it, against at least two to three weeks with Claude Code alone.
    • How ClawMetry began when OpenClaw went quiet on a side project and he asked it to build a dashboard of its own activity.
    • His roadmap: a low cost hosted version, mobile and Mac apps, and planned alerts for signs of prompt injection or attempts to delete files, with the project staying open source.
    • Why he expects lighter frameworks such as NanoClaw and PicoClaw to spread to small devices, and why products must now serve both humans and agents.

    Vivek's advice to junior engineers: "Start building whatever you wish you had."

    CHAPTERS
    00:00 Cold open
    00:23 Welcome and Vivek's path to Booking.com
    03:50 Production AI systems and eval loops
    05:28 Advice for junior engineers
    06:46 Why AI agents need observability
    09:46 Everyday OpenClaw use cases
    11:17 How much time OpenClaw saved
    12:25 What ClawMetry shows you
    14:59 How ClawMetry started
    16:38 Roadmap: hosted version, apps and security alerts
    18:39 Where agent observability is heading
    21:28 Building for humans and agents
    22:33 Vivek's AI tools and closing thoughts

    GUEST AND RESOURCES
    Vivek Chand, creator of ClawMetry
    Mentioned: OpenClaw, Claude Code, Telegram, Grafana, NanoClaw, PicoClaw, Perplexity Computer and Claude Cowork.

    FOLLOW THE NEURALPOD
    Subscribe for conversations with the founders, researchers and engineers building and applying AI.
    YouTube: https://www.youtube.com/@TheNeuralPod
    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-neuralpod/id1761461659
    The NeuralPod is hosted by Christopher Coyne, founder of NeuralRec: https://neuralrec.ai

    Recorded 5 March 2026.

    #AIAgents #OpenClaw #AgentObservability #AI #ClawMetry #AgenticAI #Observability #OpenSource #AIEngineering #AIEvals #AISecurity #ClaudeCode #AICoding #DeveloperTools #LLMOps

    24 min
  • AI Cloning Your Personality with Kate Ivanova Co-founder of Pantio

    Would you clone your voice and personality with AI so your stories could outlive you? Kate Ivanova, cofounder of Pantio, explains how the company builds digital personas from deep voice cloning and memory, and the ethical questions that come with them.

    In this episode of The NeuralPod, Christopher talks with Kate, who cofounded Pantio after three years in cancer research. She describes its aim as preserving each person's life experience, wisdom and stories for future generations. The idea grew from her cofounder creating a digital version of his late grandfather, which Kate says his grandmother now talks to for an hour every day.

    IN THIS EPISODE
    • How a persona is built: a structured interview of around three hours, deep voice cloning that captures emotion and intonation, and material users add later.
    • Why Kate sees memory as the hard problem, and why personas are designed not to simply agree with you.
    • Why Pantio is voice only for now, with holograms a longer term idea rather than video or robots.
    • Privacy and misuse: voice verification, contracts and non disclosure agreements (NDAs), encryption, no client data used in training for now, and early work on hardware that runs everything locally.
    • The grief question: the risk of unhealthy attachment, why Kate sees a persona as a softer start to grieving rather than a substitute for a person, and why she wants death doulas and other professionals involved.
    • Uses beyond legacy, from immersive biographies with caregivers to people facing terminal illness or early dementia.

    Kate also shares pricing at the time of recording: a thousand dollars per persona, including a microphone, which she says just covers the cost of serving it for a year, with prices set to rise with demand.

    Kate on what makes a persona convincing: "I think that the memory is actually a big challenge for everybody to solve."

    CHAPTERS
    00:00 Cold open
    00:30 Welcome and meet Kate, cofounder of Pantio
    02:15 What a digital persona is, and why Pantio exists
    05:29 Voice plus memory: what makes a clone convincing
    07:52 The three hour interview behind each persona
    09:25 Consumer first, with caregivers and death doulas
    10:47 The origin story: cloning a grandfather
    13:48 Misuse, privacy and what customers ask for
    17:45 Grief, attachment and unhealthy relationships
    21:54 Plans for the next six months to three years
    25:17 Holograms rather than robots
    26:35 Who Pantio is for
    29:48 How to try Pantio and what it costs
    33:18 Closing thoughts

    GUEST AND RESOURCES
    Kate Ivanova, cofounder of Pantio
    Mentioned: ElevenLabs and Delphi.

    FOLLOW THE NEURALPOD
    Subscribe for conversations with the founders, researchers and engineers building and applying AI.
    YouTube: https://www.youtube.com/@TheNeuralPod
    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-neuralpod/id1761461659
    The NeuralPod is hosted by Christopher Coyne, founder of NeuralRec: https://neuralrec.ai

    Recorded 30 October 2025.

    #AIClone #DigitalTwin #VoiceCloning #AI #DigitalLegacy #DigitalPersona #AIEthics #GriefTech #ConversationalAI #GenerativeAI #AIMemory #ElevenLabs #AIPrivacy #AIStartup #Pantio

    36 min
  • How AI is Shaping the Future of Genetics with Artem Elmuratov, Director of Research at Nebius

    How is AI changing genetics, drug discovery and gene editing? Artem Elmuratov of Nebius explains where DNA models, AlphaFold and agentic tools such as CRISPR GPT already help scientists, and what still holds the field back.

    In this episode of The NeuralPod, Christopher talks with Artem about his path from mathematics and Accenture to cofounding several genetics and medtech companies, from a DNA testing lab to confidential computing for biomedical research. They also discuss the ethics of editing human embryos, and why Artem expects attitudes to shift as understanding grows.

    IN THIS EPISODE
    • Why genetics has not yet had its "GPT moment", and how DNA, with its four letter alphabet, can be modelled much like text.
    • The challenges Artem sees: cheaper compute, a lasting shortage of good life sciences data and intense competition for talent.
    • How AlphaFold cut protein structure work from years to minutes, and why he predicts every drug that reaches the market will in some sense be AI enabled.
    • CRISPR GPT, a Stanford led agent Nebius took part in: in experiments Artem describes, molecular biologists new to gene editing designed effective CRISPR systems within a few days.
    • Why small biotech teams still need a lab for validation, and why Artem sees selling a validated molecule or technique to larger pharma early as an effective route.
    • Gattaca, embryo selection with polygenic risk scores, and how whole genome studies and omics data could lead to personalised medicine.

    Artem does not think AI will replace all scientists, adding: "But the scientists who use AI replace scientists who don't use AI."

    CHAPTERS
    00:00 Cold open: will AI replace scientists?
    00:21 Welcome: from mathematics to genetics
    03:55 Artem's career: Accenture, genetics startups and confidential computing
    05:35 Lessons from research and business, and why scientists should use AI
    08:11 Genetics and AI today: modelling DNA like language
    10:04 The biggest challenges: compute, data and talent
    12:03 AI in life sciences: AlphaFold, clinical trials and AI enabled drugs
    18:02 CRISPR GPT: an AI copilot for gene editing
    22:50 Small biotech teams, lab validation and agentic AI
    27:24 Gene edited babies and the ethics of CRISPR
    31:55 Gattaca, embryo selection and polygenic risk scores
    34:44 Longer, healthier lives and personalised medicine
    38:14 Sci fi futures: cloning, printed organs and digital copies
    40:51 Closing thoughts

    GUEST AND RESOURCES
    Nebius: https://nebius.com
    Webinar with the Le Cong Lab at Stanford: https://nebius.com/events/webinar-crispr-gpt
    CRISPR GPT paper: https://arxiv.org/abs/2404.18021
    Also mentioned: AlphaFold, the film Gattaca and the series Black Mirror.

    FOLLOW THE NEURALPOD
    Subscribe for conversations with the founders, researchers and engineers building and applying AI.
    YouTube: https://www.youtube.com/@TheNeuralPod
    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-neuralpod/id1761461659
    The NeuralPod is hosted by Christopher Coyne, founder of NeuralRec: https://neuralrec.ai

    Recorded 14 October 2025.

    #Genetics #CRISPR #AI #GeneEditing #CRISPRGPT #AIForScience #DrugDiscovery #AlphaFold #LifeSciences #Biotech #Genomics #AIAgents #PolygenicRiskScores #PersonalisedMedicine #Nebius

    43 min
  • Everything Is Prediction, Superintelligence and AI Strategy with Keith Dear PhD, Founder at Cassi

    What does it mean to treat every decision as a prediction, and how should leaders and governments prepare for artificial general intelligence (AGI)? Keith Dear, CEO and cofounder of Cassi, explains how forecasting science can make strategy measurable, and why he thinks the UK is not yet taking AGI seriously enough.

    Keith spent 18 years as an intelligence officer in the Royal Air Force, completed a PhD in experimental psychology at Oxford, advised Number 10 on science and technology and led Fujitsu's Centre for Cognitive Technologies before leaving in March 2025 to set up Cassi. He describes it as a super strategy engine: you state the outcome you want, and it uses AI and crowdsourced forecasts to estimate your odds and the factors that matter most. He also shares impressions from a UN roundtable on AI and security in Beijing.

    IN THIS EPISODE
    • Why Keith treats decisions, strategies and organisational design as forms of prediction.
    • How Cassi scores forecasters, and what happened when its AI and a customer's team disagreed.
    • What Metaculus forecasts suggest for weak AGI, Oracle AGI and superintelligence, and why definitions matter.
    • Baselining AI against human error, from hallucinations to self driving cars.
    • Why job displacement worries him most, and the risk of one country or company reaching AGI first.
    • Servant leadership, and why making the final call feels so different from giving advice.

    "Narrative is still a part of leadership, but getting to what your narrative is should be an exercise in truth seeking."

    CHAPTERS
    00:00 Cold open
    00:22 Meet Keith Dear and Cassi's super strategy engine
    04:16 From the RAF and Number 10 to founding Cassi
    14:28 What Keith learned about AI in China
    18:26 Servant leadership from the military to a startup
    26:38 When will we reach AGI? Forecasts and definitions
    32:52 The jagged frontier, hallucinations and trust
    37:54 AGI, jobs and the UK's place in the world
    42:13 Who reaches AGI first, and what Number 10 should do
    48:02 Everything is prediction
    50:50 Cassi with customers: trade offs, narrative and scoring
    59:38 Organisational design, innovation theatre and why now
    1:03:00 Self driving cars, book picks and Cassi's ambitions

    GUEST AND RESOURCES
    Keith Dear PhD, CEO and cofounder of Cassi
    Cassi: https://www.cassi-ai.com
    Keith's blog: https://blog.cassi-ai.com
    Mentioned in this episode:
    • Deep Utopia, Nick Bostrom
    • Superforecasting, Philip Tetlock and Dan Gardner
    • Servant Leadership, Robert K. Greenleaf
    • Ethan Mollick's work on the jagged frontier of AI
    • Nature Computes Better, a programme from ARIA (the Advanced Research and Invention Agency)

    Keith speaks in a personal capacity; his views are not those of the Ministry of Defence.

    FOLLOW THE NEURALPOD
    Subscribe for conversations with the founders, researchers and engineers building and applying AI.
    YouTube: https://www.youtube.com/@TheNeuralPod
    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-neuralpod/id1761461659
    The NeuralPod is hosted by Christopher Coyne, founder of NeuralRec: https://neuralrec.ai

    Recorded 3 November 2025.

    #AGI #AIStrategy #Superforecasting #AI #Superintelligence #DecisionMaking #Forecasting #AIPolicy #AISafety #FutureOfWork #NationalSecurity #ArtificialIntelligence #UKAI #ServantLeadership #AIStartups

    1 hr 8 min
  • How Agentic AI is Shaping e-commerce with Alberto Lumbreras PhD, Staff Researcher at Criteo

    How will AI shopping agents change ecommerce, product recommendation and advertising? Alberto Lumbreras of Criteo explains what agents need from retailers and their catalogues, and why he sees unique data, not models, as the real competitive moat.

    In this episode of The NeuralPod, Christopher talks with Alberto, a Staff Researcher at Criteo at the time of recording, about his route into AI: from telecommunications engineering at Telefónica, through philosophy and the study of social movements in Spain, to a PhD in France on online behaviour and machine learning.

    IN THIS EPISODE
    • How Criteo is approaching agents: aiming to be the recommender that shopping agents call, and automating client onboarding to cut cycles from days to hours or minutes.
    • Why an agent is not impressed by a display ad, and why Alberto still expects traditional ads and monetised agent spaces to coexist.
    • What changes when recommendation becomes conversational: clarifying questions, episodic and long term memory, and adapting as preferences change.
    • The tug of war between AI platforms building their own checkouts and retailers who want to control the experience, and why he sees data marketplaces as part of the answer.
    • Why, in his view, models are increasingly replicable, leaving private data such as shopping patterns, synthetic data and richer product descriptions as the real edge.
    • His advice for researchers and engineers: stay curious, use coding agents to build end to end, and keep the judgement to review their output, acting more as an architect than a developer.

    CHAPTERS
    00:00 Cold open: what changes and what never does
    00:19 Welcome: from telecoms engineering and philosophy to AI research
    04:18 How shopping agents will change ecommerce and ad tech
    10:25 Beyond classic recommenders: conversation, memory and adaptability
    14:10 Adoption, agentic checkout and data marketplaces
    20:26 Why data, not models, is the moat
    24:18 Overlooked data: product descriptions, travel and user memory
    28:21 What ML researchers need from engineering, and coding agents
    34:00 Career advice: build end to end, and will teams shrink?
    37:28 Predictions: smart glasses, open source, small models and autonomy
    42:23 The AI tools Alberto uses, and why MCP matters
    43:59 Research to watch: tiny recursive networks and generative retrieval
    47:28 Books on digital pollution and the enclosure of data
    50:51 Closing thoughts

    GUEST AND RESOURCES
    Criteo: https://www.criteo.com
    Paper mentioned: Less is More: Recursive Reasoning with Tiny Networks: https://arxiv.org/abs/2510.04871
    Books mentioned: Caliban and the Witch by Silvia Federici, and The Many Headed Hydra by Peter Linebaugh and Marcus Rediker.

    FOLLOW THE NEURALPOD
    Subscribe for conversations with the founders, researchers and engineers building and applying AI.
    YouTube: https://www.youtube.com/@TheNeuralPod
    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-neuralpod/id1761461659
    The NeuralPod is hosted by Christopher Coyne, founder of NeuralRec: https://neuralrec.ai

    Recorded 13 October 2025.

    #AgenticCommerce #AIAgents #AI #Ecommerce #AdTech #ShoppingAgents #RecommenderSystems #DataMoat #SyntheticData #GenerativeRetrieval #OpenSourceAI #SmallLanguageModels #MachineLearning #MLResearch #Criteo

    52 min
  • From ByteDance RecSys to Vertical AI for Legal with Michal Klos, Co-Founder at Lex AI

    What does it take to build AI that legal teams will trust? Michal Klos, cofounder of Lex AI, explains how his team turns a constant flow of regulatory change into a searchable legal radar, with human review before anything is published.

    In this episode of The NeuralPod, Christopher talks with Michal about a career that began in video game development and led to ByteDance, where he worked on real time graphics and on device machine learning (ML) for creative effects, and then to ShareChat, where he optimised the cost of the recommender system pipeline. He joined Lex AI after it was founded, drawn by its focus on gathering data before generating anything.

    He also explains why, in his experience, lawyers in Germany expect especially high accuracy from AI, and why he counts the Deloitte partnership among the team's biggest recent wins.

    IN THIS EPISODE
    • Why Michal sees data, not model training, as most of the work in ML, and when chasing a fraction of a percent pays off.
    • How the Lex AI pipeline works, from automated workflows for each authority to combined full text and semantic search.
    • Why mapping a team's workflow, and its internal politics, should come before adding agents.
    • Why lawyers asked to see the AI, and the chatbot Michal built between a Friday and a Monday in response.
    • Why big tech habits, such as building perfect systems in house, can slow a startup whose priority is runway.
    • How Michal manages energy as a founder, and how he uses ChatGPT and coding assistants.

    CHAPTERS
    00:00 Welcome: from programming and video games to ML
    03:44 ByteDance: real time graphics and on device ML
    06:37 ShareChat: low end devices and recommender system costs
    09:45 ML fundamentals: why data comes first
    15:30 What an MBA adds, and the big tech to startup mindset shift
    23:57 Lex AI: a legal radar for regulatory change
    27:30 Why legal teams resist change, and mapping workflows before agents
    32:18 Under the hood: search, human review and hallucination risk
    39:46 Building trust, the Deloitte partnership and who Lex AI serves
    43:21 Product market fit and a chatbot built over a weekend
    54:57 Do big tech engineers over engineer at startups?
    1:00:27 Where legal AI is heading in Europe and the US
    1:04:32 Founder energy, AI tools and favourite books
    1:15:36 AI trends, generative video and closing thoughts

    GUEST AND RESOURCES
    Lex AI: https://lexai.co
    Creativity, Inc. by Ed Catmull
    Joel Spolsky's stories from Microsoft
    Tools discussed: PostgreSQL, Elasticsearch, Hugging Face, ChatGPT

    FOLLOW THE NEURALPOD
    Subscribe for conversations with the founders, researchers and engineers building and applying AI.
    YouTube: https://www.youtube.com/@TheNeuralPod
    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-neuralpod/id1761461659
    The NeuralPod is hosted by Christopher Coyne, founder of NeuralRec: https://neuralrec.ai

    Recorded 11 September 2025.

    #LegalAI #LegalTech #AI #VerticalAI #RegTech #RegulatoryCompliance #MachineLearning #HumanInTheLoop #SemanticSearch #ProductMarketFit #StartupFounder #MLEngineering #ByteDance #GenerativeAI #AIStartups

    1 hr 24 min
  • Organ Replacement and Blood Boys, how the Super Rich Live Longer with Alexey Strygin, Viva City

    Could replacing organs, rather than curing ageing directly, be the faster route to longer lives, and how far can AI help? In Part 2 of AI and Longevity, Alexey Strygin, a cofounder of Viva City, discusses organ replacement research, the case for a city built around longevity science and the AI Agents vs Aging hackathon.

    Returning to The NeuralPod, Alexey tells Christopher that progress is still slow: in his view no drug is yet proven to extend human lifespan, and, by his figures, the field receives less than 0.1% of total billionaire net worth. He describes a proof of concept he saw at the ARDD (Aging Research and Drug Discovery) conference: a mouse sac of organs grown without a brain or central nervous system.

    Viva City aims to secure a special economic zone whose tax and legal rules would speed up drug discovery, inspired by megaprojects such as Apollo, the Manhattan Project and the International Space Station. Alexey also gives a reality check on claims that world leaders could live to 150, and explains why he would not rely on AI alone to solve ageing.

    IN THIS EPISODE
    • Why Alexey sees 3D bioprinting as a likely dead end, why growing organs may be cheaper, and where the ethical lines on sentience sit.
    • How safer self driving cars could deepen the organ shortage, since many donor organs come from people who die in car accidents, and where gene edited pig organs fit in.
    • Why Viva City is negotiating with governments for land and legal freedom, and why being near hubs such as the Bay Area or Boston matters.
    • Ideas for early access to therapies without repeating the stem cell industry's mistakes, including insurers acting as regulators and payments released only once a therapy is shown to work.
    • What lifestyle can offer today, about 10 years on his reading of the papers, and his fear that his generation could be the last to die.
    • The two tracks of the hackathon he was organising at the time of recording, one curated with Retro Biosciences.

    "It's not enough just to sit on the couch and see how the wonderful longevity future unfolds." Alexey Strygin

    GUEST AND RESOURCES
    Alexey Strygin, Viva City
    Viva City: https://viva.city/
    Part 1 with Alexey is a separate episode of The NeuralPod.

    Mentioned in this episode
    • ARDD, the Aging Research and Drug Discovery conference
    • Retro Biosciences
    • Viva Frontier Tower, Viva City's pop up community in San Francisco
    • Montana's right to try law for early access to therapies

    FOLLOW THE NEURALPOD
    Subscribe for conversations with the founders, researchers and engineers building and applying AI.
    YouTube: https://www.youtube.com/@TheNeuralPod
    Apple Podcasts: https://podcasts.apple.com/us/podcast/the-neuralpod/id1761461659
    The NeuralPod is hosted by Christopher Coyne, founder of NeuralRec: https://neuralrec.ai

    Recorded 23 September 2025.

    #Longevity #OrganReplacement #AI #LongevityResearch #Biotech #AIAgents #Hackathon #SpecialEconomicZones #LifeExtension #DrugDiscovery #Xenotransplantation #OrganShortage #HealthyAgeing #VivaCity #Bioethics

    59 min