Vivek Vishwakarma is an entrepreneur, investor, and technologist leading ThirdAI Automation, an industrial AI company that accelerates troubleshooting and reporting through automated root-cause analysis. A former technologist at Intel with 10+ patents and 300+ research citations, he speaks on the intersection of advanced manufacturing, data, and AI.
Tell us about your company.
ThirdAI Automation is an industrial AI company building Causal AI for the semiconductor equipment ecosystem. Co-founded by Vivek Vishwakarma (CEO) and Dr. Sainyam Galhotra (CTO and Cornell University faculty, with deep research credentials in causal inference), the company is backed by a $3M seed round led by Endiya Partners and Capria Ventures. We operate across dual headquarters in San Francisco and Bengaluru, with active pilots across leading semiconductor equipment vendors and select Fortune 500 manufacturers, spanning the U.S., Japan, Israel, and Taiwan. Our platform converts fragmented logs, sensor telemetry, and tribal engineering knowledge into automated root-cause intelligence, engineered to run on standard CPU infrastructure inside the air-gapped, on-premise environments where our customers’ tools operate.
What problems are you solving?
We solve the diagnostic-latency problem that sits between equipment vendors and their fab customers. When a CMP, lithography, etch, deposition, or packaging tool experiences a fault or process excursion at a customer site, the equipment vendor’s field-service and applications-engineering teams typically spend hours to days digging through logs, recipes, and historical incidents to isolate the root cause, while the customer’s tool sits down and wafers accumulate at risk. Our Causal AI agents automate this work, compressing diagnostic cycles that historically take 8 hours of expert time down to roughly 12 minutes, and surfacing not just what failed but why. The economic stake is substantial on both sides: tool vendors protect their uptime SLAs, lower field-service costs, and codify the tribal expertise of senior engineers; their fab customers avoid the millions in scrapped wafers a single calibration drift can cause at advanced nodes.
What application areas are you strongest in?
Our strongest position is across the semiconductor equipment vendor ecosystem, spanning Front-End-of-Line (FEOL), Back-End-of-Line (BEOL), and advanced packaging. We work with vendors of CMP, lithography, etch, deposition, metrology, inspection, bonding, and dicing tools, where unplanned downtime at a customer fab carries the highest cost. Our deployments are designed to plug into the vendor’s own service workflow, supporting field-service engineers, applications engineers, and process-development teams, and increasingly inside the equipment itself, so the diagnostic intelligence travels with the tool to the customer site.
What keeps your customers up at night?
Tool vendors today are squeezed on multiple fronts. Uptime SLAs at advanced nodes are tighter than ever, and a single delayed root cause can trigger penalty clauses or, worse, escalate into a customer relationship issue. Field-service costs keep climbing, sending senior engineers on-site to diagnose intermittent issues is expensive, slow, and increasingly hard to staff as veteran engineers retire and take decades of tribal knowledge with them. On top of that, fab customers won’t allow cloud connectivity into their environments, which rules out most modern AI options for vendors trying to scale diagnostics across their installed base. Our deployment model addresses all of this: faster RCA, codified institutional knowledge, and CPU-based on-premise inference that fits inside the most secured customer environments.
What does the competitive landscape look like, and how do you differentiate?
The market sits between two extremes: legacy SPC and FDC tools that detect anomalies but don’t explain them, and modern AI vendors (Augury, UptimeAI, Aitomatic, GaussLab/PDF Solutions, Ethon.ai, CausalLens) that require heavy GPU infrastructure and often cloud connectivity that tool vendors simply cannot deploy inside their customers’ fabs. We differentiate on three axes. First, causality: most AI tells you what happened, our platform tells you why, anchored in our CTO’s published research in causal inference, which is what closes the diagnostic loop rather than producing more alerts. Second, infrastructure fit: our engine runs on standard CPUs and is built to ship inside on-premise, air-gapped customer environments , exactly where our equipment-vendor customers need to land. Third, time-to-value: structured pilots typically deliver measurable RCA acceleration within 8 to 12 weeks of ingest.
What new features and technology are you working on?
We are extending our Automated Root Cause Analysis (ARCA) agents in three directions. First, multi-modal data ingestion, combining structured machine telemetry with unstructured sources like maintenance manuals, ECN notes, and engineer ticket logs. Second, closed-loop diagnostics, where the platform not only identifies the failure but also surfaces the precise maintenance or recipe-adjustment protocol from the tool’s documentation, turning every diagnostic event into an action. Third, we’re rolling out a dedicated MLOps platform that allows equipment vendors and their teams to manage their own models, including bring-your-own-model workflows, within the same lightweight inference layer, so vendors can extend the platform into their own intellectual property over time.
How do customers normally engage with your company?
Engagement typically begins with a structured pilot, 8 to 12 weeks, where we ingest historical log and telemetry data from a specific toolset and benchmark RCA performance against the customer’s existing process. Successful pilots transition into a multi-year platform license, deployed either inside the customer’s private cloud or fully on-premise within their service infrastructure. Pricing is tiered (Pilot, Professional, Enterprise) with hybrid subscription plus usage components. For equipment vendors who want the diagnostic intelligence to travel with their tools to end customers, we also offer a per-tool OEM licensing model, allowing vendors to embed the platform inside the equipment they deliver, while preserving customer data sovereignty and ThirdAI’s underlying platform IP.
Also Read:
CEO Interview with Vivek Raghunathan of Xscape Photonics
CEO Interview with Baratunde Cola of Carbice
CEO Interview with RP Singh of Seasia Infotech
Share this post via:
Disaggregating AI Compute to Break the Tokens Barrier