A circuit diagram is a tidy thing. Lines meet symbols. Components sit obediently in rows. The page gives no hint that one transistor can unsettle an entire design, that a simulation may approve a chip reality later rejects, or that the fix might live only in the memory of the engineer who has been stung by this exact failure before. Sayan Mitra has built his young career around the distance between a clean representation and a physical result.
Now he is co-founder and CEO of Atrisa, a San Francisco startup in Y Combinator's Spring 2026 batch. Its brief is unusually precise: build AI agents that help analog engineers design circuits, run simulations and investigate failures. The product sits beside the engineer through the loop rather than waiting at the edge of it with a generic answer.
That specificity has a biography behind it. Mitra studied electrical engineering at IIT Madras from 2018 to 2023. He worked on the institute's Avishkar Hyperloop team, held a Young Research Fellowship, and took AI and machine-learning research roles at the Indian Institute of Science and IBM Research. In 2022 he interned at KLA, the semiconductor equipment company, then stayed as an artificial intelligence engineer through 2025. By the time he founded Atrisa with Atman Kar and Rithik Jain, the two halves of the problem were already on his resume: circuitry that must obey physics, and software that learns through examples.
The stubborn half of the chip
Digital design benefits from layers of abstraction. Ones and zeroes are obliging that way. Analog design lives nearer the unruly world of voltage, current, temperature and manufacturing variation. Change one component and parasitics or interference can alter the behavior of the whole. Debugging becomes a conversation with a system that answers indirectly.
This is why analog expertise is often described as intuition. An experienced designer recognizes patterns assembled across years of experiments, failed layouts and late-night investigations. The knowledge is valuable precisely because it was expensive to acquire. It is also difficult to turn into a manual. A junior engineer can read the documentation and still need a senior colleague to say, in effect, "I have seen this shape before."
“Dreams are just synthetic data that your brain RL trains you on.”Sayan Mitra, in a public post
Mitra's line about dreams is a joke, but it contains his professional lens in miniature. Experience becomes data; feedback changes behavior; intuition is the compressed result. Atrisa applies a similar logic to circuit work, with a crucial advantage over many AI applications. Circuits have verifiers. A SPICE simulation can score a design against voltage, power, gain or stability requirements. The answer must do more than read plausibly. It has to survive the test bench.
Atrisa's public workflow begins with a design goal and moves through schematic changes, analysis and debugging. The system is meant to absorb existing toolkits, process design kits and design documents rather than asking an engineering team to abandon its working environment. It can reuse older designs and account for layout context. The promise is continuity: fewer manual handoffs, less repeated setup, and a record of the reasoning that brought a circuit from intent to result.
A founder shaped by research loops
The sequence of Mitra's early work matters. IISc, IBM Research and KLA are different institutions, but each put him near problems where an idea has to be tested. His IBM internship focused on natural-language processing. KLA placed AI inside the semiconductor world. Avishkar Hyperloop, meanwhile, was the sort of student project where theory eventually meets a machine that either moves or declines to cooperate.
There is another thread, less technical but just as revealing. In January 2021, while a third-year undergraduate, Mitra helped create a team of ten IIT Madras students to mentor twenty children across three Shiksha Prayas centers. He led the mentor team. In a volunteer diary written that May, he described the satisfaction of connecting with students personally, guiding them through academic and non-academic questions, and streamlining the process around them.
Ten university students. Twenty younger students. Three centers. Mitra's role was not only to mentor, but to assemble a system through which more mentoring could happen.
It would be too neat to call that a rehearsal for a startup. Still, the family resemblance is hard to miss. Both projects begin with knowledge locked inside individuals. Both ask how that knowledge can travel without losing its usefulness. Both preserve a role for judgment while improving the machinery around it.
His public voice also resists corporate varnish. He has joked about LinkedIn broadcasting every small action to his network, complained colorfully about a rainy Diwali in Bangalore, and recently posted about Gödel's letter to John von Neumann and the idea of a proof machine. The range is appealingly human: social-platform irritation, weather, dreams as reinforcement learning, then computational complexity. Curiosity does not always arrive wearing a conference badge.
The company after the pivot
Atrisa was previously called Refortif.ai and worked on protecting deployed AI model weights. During the team's Y Combinator period, it pivoted toward analog circuit design. A co-founder later described those months as a visit to "Pivot Hell," followed by the arrival of Atrisa. The new direction joined the founders' complementary backgrounds: Mitra's AI and electrical engineering, Kar's experience at Texas Instruments, and Jain's high-performance computing work.
The product they launched in June 2026 reasons over circuits, proposes and evaluates changes, and helps investigate verification failures. Its architecture uses multiple agents because a single linear explanation does not resemble the branching way an expert troubleshoots. One possibility leads to an experiment; the result changes the next possibility. The startup has also emphasized repeatability, an essential quality in a field where a charmingly creative answer can become a very expensive piece of silicon.
The startup opportunity is hiding in the sentence an experienced engineer begins with: “Usually, when I see this...”
Training material presents another constraint. Companies closely guard circuit designs, and much of the useful analog record never reaches the public internet. Atrisa has said it built a custom collection of roughly 2,000 circuits. That is small beside internet-scale model corpora and large enough to suggest the nature of the job. This is not a sweep across all human text. It is the careful construction of a technical pattern library, connected to simulators that can push back.
Software meets silicon
Atrisa's immediate product is a copilot. The longer arc is more agentic: a system able to carry larger sections of the design loop while engineers set direction and judge tradeoffs. Layout and post-silicon debugging remain frontiers. So does trust. An analog engineer takes professional pride in finding the brutal little bug, and will not surrender a workflow merely because a chat box has learned the vocabulary.
Mitra's challenge as CEO is therefore cultural as well as technical. Atrisa must show its work, fit the tools already in use and prove value in a community where reputations travel quickly. Its website makes a deliberately workmanlike argument: tell the system what to build, change or verify; let it manage repeated setup and tool switching; keep the engineer focused on judgment.
This is the quieter meaning of the company's claim that one engineer can carry the capacity of a larger team. More output is useful. Shared memory may be more important. If an experiment, failure and fix remain available to the next engineer, the team's experience compounds instead of vanishing into a meeting or a senior designer's recollection.
It also changes what a junior engineer can ask. Instead of waiting for the one colleague who remembers a similar failure, the engineer can begin with the circuit's accumulated record, test a hypothesis and bring a sharper question to the team. Senior judgment is still scarce. Better preparation lets a company spend that scarcity on decisions, not retrieval.
Mitra has reached this point quickly: graduate in 2023, semiconductor AI engineer, founder in 2026. The company is younger still. Its public evidence is a product, a demo and an exacting thesis, not a long corporate history. Yet the question beneath it is mature. How do you make experience transferable without pretending it is simple?
Analog design is a fine place to ask. Physics is patient, literal and indifferent to presentation. A circuit either meets its constraints or supplies fresh evidence. For a founder with one foot in AI and the other in electrical engineering, that makes the field difficult in precisely the productive way. The machine may learn the instincts. Silicon will grade the exam.