The modern semiconductor factory has the choreography of science fiction. Wafers travel through sealed rooms. Robotic arms make precise movements. Hundreds of machines record signals across thousands of steps. Then, in the middle of all that automation, a person may still need to look at defect images and decide what went wrong.
Akanksha Jagwani built her company around that contradiction. SixSense, which she co-founded with Avni Agrawal in Singapore in 2018, gives factory engineers software for classifying defects, spotting failure patterns, and tracing trouble back through a production line. Its promise is practical: find the weak signal before it becomes a costly pile of rejected wafers.
Jagwani did not arrive with a lifetime inside a fab. She came with a mechanical engineering education, experience in manufacturing automation, and a product designer's habit of watching how people actually work. That mix matters. Industrial AI is full of technically impressive models that struggle to cross the last few feet between a demonstration and a production decision. SixSense lives in those feet.
A mechanical engineer learns to look twice
At IIT Gandhinagar, Jagwani studied mechanical engineering from 2010 to 2014 and appeared repeatedly on the Dean's List. Her projects had the texture of physical systems: thermal models for low-cost cooling in hot, dry climates; simulations of ballistic impact; a computational model for improving a micro-wind-turbine blade. During a 2012 summer research experience at Washington University in St. Louis, she joined a lab working on energy systems.
She also behaved like someone interested in more than equations. She co-founded Udbhav, an educational initiative that brought together 30 volunteers to encourage scientific thinking and self-directed learning among secondary-school students. She served as secretary of the Arts Club and helped organize touch rugby. Years later, her company biography would mention dancing and experiments with food as the ways she steps away from the fab.
After college, her path moved from engineering toward the interface between systems and users. At Altair Engineering she supported customers including Hyundai Motors and General Electric. She later worked in product roles at education company Embibe and travel startup Headout. The sequence supplied two kinds of fluency: how complex machinery behaves and how a useful product fits into a human workflow.
Could we help semiconductor fabs see what even the sharpest human eye missed?Akanksha Jagwani, recalling SixSense's founding question
The one-week partnership
The origin of SixSense is less tidy than a rehearsed founder story. Jagwani and Agrawal met through Entrepreneur First's Singapore program in 2018. Agrawal was a computer scientist who had worked with large data systems at Visa and D. E. Shaw. Jagwani understood manufacturing and product. They shared an interest in applying machine learning to an industry where small process failures produce large economic consequences.
They decided to work together one week before the program's final investment pitch. The pair worked through the night to assemble the plan. Their first pitch did not make the cut. Yet the program saw enough in their determination to grant an unusual month-long extension and convene another panel. On the second attempt, they secured a $75,000 pre-seed investment. SixSense formally began in December 2018.
The valuable detail is what happened next. The founders did not treat “manufacturing AI” as a sufficient idea. They consulted more than 50 engineers, many from semiconductor companies, and listened for recurring friction. The inspection gap kept returning. Highly automated factories were still asking trained operators to review enormous volumes of images, classify defects, and decide which chips could continue. Human judgment remained essential, but speed, consistency, and scale were under pressure.
From wafer image to factory decision
A simplified view of the SixSense workflow
The trust layer is part of the product
A fab does not adopt a model merely because its accuracy looks good on a slide. A wrong call can scrap a usable chip, miss a real defect, or allow a process drift to continue across a lot. Engineers need to understand what the system saw. They need to retrain it when materials, devices, and defect types change. They need it to work with existing inspection hardware and fit the cadence of production.
SixSense designed around those constraints. Its platform is hardware-agnostic, explainable, and usable without code. Factory engineers can tune models with their own data instead of shipping every change to a separate data-science team. The company has said deployment can happen in under two days, while configurations for some image and defect types can be set up in hours. The practical point is control: the people accountable for yield can see, challenge, and improve the model.
Jagwani leads business development and works with fabs around the world on deployment. This is an unusually technical sales environment. The buyer is not purchasing a generic productivity tool. The conversation touches production history, defect taxonomies, tool interfaces, data governance, and the internal politics of changing a quality decision. Trust is earned in small increments, then tested at line speed.
What customers report
Selected SixSense customer outcomes reported in 2025
From classification to anticipation
The company's first public wedge was automated defect classification. Its classifAI software learned visual patterns across inspection layers, devices, and technology nodes, then classified new production images at high speed. Early adopters included Infineon Technologies and GlobalFoundries. By 2025, SixSense named GlobalFoundries and JCET among its customers and said the platform had processed more than 100 million chips.
The ambition has widened along a logical path. Classification answers, “What is this defect?” Root-cause analysis asks which tool, step, or mix of process conditions produced it. Prediction asks where the same pattern may appear next. Each step moves the factory earlier in time, from sorting the consequences toward controlling the cause.
Jagwani has called quality and yield one of the underloved layers of AI infrastructure. The observation lands because every argument about more compute eventually encounters manufacturing reality. A chip design has no economic life until enough working units emerge from the line. Yield is where microscopic variation becomes supply, margin, and delivery.
The longer you spend in semiconductors, the smaller the industry feels. The work compounds, and so do the relationships.Akanksha Jagwani after ASMC 2026
A second chance becomes a corridor
SixSense grew along the geography of chipmaking. It began in Singapore, built a team across Singapore, India, and Taiwan, and worked with manufacturers in Asia, Europe, and the United States. Its 2025 Series A, led by Peak XV's Surge with Alpha Intelligence Capital, FEBE, and other investors participating, gave the company $8.5 million to deepen research and expand in Malaysia, Taiwan, and the US.
In 2026, the World Economic Forum named SixSense one of 100 Technology Pioneers from 23 countries. Jagwani used the moment to credit the team, early customers, her co-founder, and the investors who backed what she described as an important but unglamorous problem. Soon after, SixSense started recruiting a founding US pre-sales hire to build its American go-to-market operation.
Mechanical foundations
IIT Gandhinagar, engineering research, product prototypes, and repeated Dean's List recognition.
A late pairing and a second pitch
Jagwani meets Agrawal at Entrepreneur First. SixSense begins after a one-month extension.
Into production
ClassifAI is documented with major semiconductor manufacturers as SixSense expands across regions.
Capital and recognition
An $8.5 million Series A funds expansion, followed by selection as a WEF Technology Pioneer.
What founders can steal
Jagwani's story offers a crisp product lesson. Find the manual judgment hiding inside an automated system. Learn who carries the cost of a bad decision. Build the model, interface, and feedback loop around that person's reality. In SixSense's case, the important user is the engineer who has to trust a classification at two in the morning while a production line keeps moving.
There is also a lesson in sequence. The founders began broad, talked to 50-plus engineers, chose a specific bottleneck, and earned the right to expand from classification into prediction and process control. The narrow entry did not limit the vision. It made the vision operational.
Outside work, Jagwani says she dances and experiments with food that pairs nutrition with taste. It is a warm detail in a career filled with microscopic defects and manufacturing data, but it also fits her method: test a combination, notice what works, adjust, repeat. Inside the fab, that loop is becoming software. The factory looks, learns, and gets another chance to catch the trouble early.