Madhulima Tewari is the CEO and founder of VerifAIX. After two decades building EDA tools, taping out chips, and working on distributed systems, cloud infrastructure, and enterprise AI applications, she teamed up with chip design and methodology leaders and university researchers to build an agentic AI system aimed at helping the industry build provably correct chips. The founding team brings mathematical proof and neuro-symbolic techniques to the platform.
Tell us about your company
Chip design and verification have no margin for error, so as AI takes on a larger role in the design workflow, the central issue becomes trust: not just how fast a team can design, but how much they can trust what they’ve designed.
VerifAIX’s spec-first agentic AI prioritizes the creation of a correct specification and understanding of design intent, using that foundation to build a trust layer which then drives automation of verification plans, testbenches, assertions, and coverage workflows. This approach combines the strengths of modern LLM reasoning with the mathematical precision of deterministic algorithms, delivering provable correctness and full traceability across the flow. This verification trust layer for chip design, a Formal Brain behind the AI, is the key to what VerifAIX brings to its customers.
The problem VerifAIX is solving
Verification remains one of the biggest bottlenecks in chip design, and even with all the time invested, bugs and gaps still creep through, much of it in test plans, assertions, testbenches, simulation, and debug that are still built largely by hand.
Many new AI verification tools and solutions now promise to speed this up, but speed alone does not solve the underlying problem. AI that generates and checks its own output, with no independent grounding, can move fast without being provably correct. Additionally, the specification a team is verifying against is often incomplete or simply wrong, so even careful verification can end up confirming a chip against the wrong intent.
VerifAIX’s verification trust layer is built to address all of these issues. It grounds the AI’s output in deterministic, mathematically precise analysis, so every result can be traced through each step of the verification flow. It also works across the RTL, existing specs, and tests together to get the specification right in the first place. VerifAIX can then automate the creation of test plans, assertions, and testbenches, grounded in a correct specification and backed by traceable, deterministic results. The result is increased verification productivity, reduced verification cost, improved quality and correctness, and accelerated time to silicon with confidence.
What application areas are your strongest
The company’s strongest area is functional verification for control-intensive IPs and subsystems, designs where correctness depends on complex protocols, state machines, ordering rules, error handling, resets, interrupts, and countless corner cases. Examples include interconnects, control and management blocks, address translation, coherency-related logic, and subsystem control paths.
It’s a natural fit for the platform’s approach: reasoning directly from specifications, extracting design intent, and producing verification artifacts that normally take expert engineers weeks to build by hand. VerifAIX’s sweet spot is specification-heavy, stateful logic, where manual verification is hardest, slowest, and most dependent on scarce expertise. To manage the complexity of modern designs and protocols, the platform uses automated decomposition and abstraction to break large verification problems into smaller, tractable tasks while preserving traceability to the full design behavior and specification intent.
What keeps your customers up at night
Two major concerns keep customers up at night: verification confidence and tape-out schedule. Even with months of verification and millions of CPU-hours, teams still wonder whether they’ve verified enough, and every additional week spent debugging or closing coverage can delay tape-out.
As the use of AI in the design and verification process grows across the industry, a newer worry is emerging: teams now need the right tools to trust what the AI produced, not just what their engineers wrote by hand.
What VerifAIX sees across customer engagements is that these three concerns, verification confidence, tape-out schedule, and trust in the use of AI, do not exist in isolation from one another. Customers come to VerifAIX because addressing these together, rather than one at a time, is what makes tape-out schedules more predictable and verification rigor measurable.
There are many AI companies and solutions. How are you different from them?
General-purpose coding and agent companies are pushing into hardware, EDA incumbents are adding AI to existing verification tools, and startups are building point solutions for specific verification tasks. Most are converging on agentic architectures, and VerifAIX builds on that same foundation.
But an agent is only as good as its ability to verify its own output. Most agentic systems still resolve to an LLM’s judgment at the critical moment, fluent, but not deterministic. Piling on more checkers and more agents does not solve that.
What makes VerifAIX different is where the ground truth sits: its trust layer. Rather than asking the model to be right, VerifAIX checks its output against mathematically grounded methods that either hold or they do not, and records the result as reproducible evidence.
That neuro-symbolic combination is the rigor underneath what customers actually sign a chip off on, not the model itself. Better models will make the VerifAIX platform stronger, but that foundation is the trust layer, not the frontier model.
What new features/technology are you working on?
VerifAIX is expanding on two fronts: the surface it covers, and the depth of the trust underneath it. On coverage, the company is moving from complex block verification toward system-level and hardware-software verification, helping customers verify how blocks interact across interconnects, memory, and full system scenarios, and that hardware and software work correctly together, not just in isolation. Since formal methods can’t handle a full chip, the platform analyzes a design to identify which parts are strong candidates for formal verification up to the subsystem or IP level, and routes the rest to simulation at the full-system level, driving both jointly.
On depth, the team is strengthening the trust layer itself, along with sharper root-cause analysis and end-to-end provenance, so every result carries reproducible evidence. As systems get larger, the rigor underneath has to scale with them. The goal is confidence not just that each block is correct, but that the whole system works, with the evidence to prove it.
Semiconductor companies are developing their own agentic flows. How does VerifAIX fit in?
EDA workflows are undergoing an unprecedented transformation. It’s natural that semiconductor companies are building their own agentic flows, and that’s exactly the world VerifAIX is built for.
VerifAIX is the trust layer inside the flows. It integrates seamlessly using industry-standard protocols like MCP, so its verification agents plug into whatever platforms, coding agents, and orchestration frameworks a customer already uses.
A customer’s agentic flow can be fast, but speed alone does not establish whether its output is correct. That requires a source of ground truth, something that determines correctness and provides evidence to back it, which is what VerifAIX provides. The more agentic a customer’s flow becomes, the more it needs a trust layer it can rely on, and VerifAIX supports that move into a multi-agent environment with correctness that can be proven.
How do customers normally engage with your company?
Customers typically engage with VerifAIX in one of two ways, depending on how much hands-on support they want. The first is platform licensing, where customers get VerifAIX licenses and their own design and verification engineers operate the platform and agents directly, with standard onboarding and integration support. The second is a higher-touch model that adds a Forward Deployed Engineer, a verification expert who works directly with the customer’s design and verification teams, drives integration, configuration, and methodology setup, and helps run the VerifAIX tools directly on the customer’s design.
Where can we meet VerifAIX
Readers can learn more about VerifAIX or get in touch directly at www.verifaix.com. VerifAIX will also be at DAC 2026 in Long Beach, Booth 854, where meetings can be booked in advance at https://verifaix.com/meet-us-at-dac-2026/.
Also Read:
CEO Interview with Dr. Albert Liu of Kneron
Executive Interview with Ebrahim Hussain and Aaditya Subediand of Architect Labs
CEO Interview with Mark Ren of Agentrys
Share this post via:
Consolidation and Competition: Who is Winning the $4.5 Billion Interface IP Race?