Across client engagements and six products, Pant worked where AI strategy met field research, product design, implementation and go-to-market.
A seven-person strategy firm had a recurring problem. Before its analysts could interpret campaign results, they spent 15 to 20 hours processing each wave of data. Nitesh Pant led a DevDash Labs engagement that examined the workflow and replaced much of the manual preparation with an automated data pipeline. Processing fell to three to five hours per wave, freeing an estimated eight weeks of analyst capacity a year.
The assignment is one part of Pant's work after leaving strategy consulting in 2024. Through July 2026, he led client engagements involving research automation, privacy-sensitive enterprise systems, AI product planning and go-to-market execution. He also led product work across six applied AI products. Four did not find a market.
Nitesh Pant is an applied AI product builder focused on AI-native software for professional-services firms. His work combines product decisions about what to build with growth questions about whether buyers understand it, trust it and change behavior because of it.
A Foundation in Learning Unfamiliar Systems
Pant grew up in Dhangadhi, a city in Nepal's far-western plains. In grade ten, he read a news report about a student admitted to all eight Ivy League universities. He did not know what the Ivy League was, and nobody around him had gone through the American admissions process. He travelled to Kathmandu for preparation books and taught himself how the system worked.
Pant entered Dartmouth College in 2018 and studied economics and government. After a pandemic gap year, he completed his degree in June 2023. The method he had used from Dhangadhi later reappeared in his work: enter an unfamiliar system, identify its decisions and constraints, and work backward toward something that can be tested.
From Strategy Consulting to Applied AI
After Dartmouth, Pant joined Roland Berger's Boston office and worked on projects in the chemicals sector. One assignment required him to build cost curves by finding plant and production information in reports, extracting the figures and organizing them in spreadsheets. The experience placed him inside the analyst layer of consulting, where information is collected, checked and prepared before a senior professional interprets it.
Pant believed much of that preparation could be automated. He left the firm in 2024 and began working at DevDash Labs that June. A later proposal to automate parts of the cost-curve process for his former employer stalled over data access and contracting. It never became a project. Technical feasibility had not answered whether the system could enter the organization.
From June 2024 through July 2026, Pant's work at DevDash Labs covered product definition, user research, client delivery and go-to-market. For a manufacturing company working through an advisory partner, he led delivery of a market-intelligence system that scanned more than 50 German and English sources and processed more than 500 signals a day. It looked for procurement-specific developments and delivered a short list for human review.
Another engagement involved workforce analytics built from computer screenshots. The proposed users worked in credit unions, insurance and healthcare, where an image could contain protected information. Pant led the workshops, client alignment, product requirements and development planning. Working with the technical lead, he helped define an architecture that kept sensitive images inside the customer's controlled environment.
For a healthcare software company, Pant worked as AI product architect and project lead. Pant and the technical team reorganized three disconnected AI ideas into two products, assessed build-versus-buy options and produced a seven-phase roadmap. The technologies differed, but each assignment began with the same questions: what work consumed time, which data could move, who reviewed the output and what might stop adoption?
Growth as a Product Discipline
Pant's work also included assignments with little software development. A healthcare compliance software company had built an anesthesia-management platform for dental specialists but needed a commercial foundation before an industry trade show. Pant did not build the clinical platform. During a four-week engagement, he led its positioning, website launch, customer definition, HubSpot setup and trade-show preparation.
The engagement stopped after its first phase. Pant carried a practical connection between adoption and design into later product work. Customer interviews informed the problem definition, demonstrations exposed objections, and commercial conversations tested pricing and urgency. For Pant, a product that worked technically but could not clear those tests was unfinished.
Product Judgment Built Through Repetition
Across roughly two years, Pant led product work on six applied AI products. Four did not find a market. An enterprise conversational AI system became infrastructure for later products. Atlantis coordinated specialized agents to produce cited research. GrowthOS combined market monitoring, prospect research and outreach, but it was closed after failing to find buyers.
In January 2026, Pant published nocaap, an open-source developer tool he designed and wrote. Almost nobody outside DevDash Labs used it. The failure produced a lesson that carried into his later work: the language model was rarely the main constraint. A capable model still performed poorly when it lacked the organization's history, terminology, priorities and rules.
Across these products, Pant researched users, defined problems, wrote specifications, set priorities and worked with engineers on system behavior. Tilak Joshi led engineering, while engineers in Kathmandu handled most production implementation.
Research Before Architecture
Work on alkemy, the sixth product, began in November 2025. Pant initially expected consulting and professional-services firms to ask for better lead generation. More than 125 conversations with owners, partners, buyers and advisers pointed elsewhere. Many firms already had contact databases, but business development depended on one or two senior people who carried years of relationship history in their heads.
Pant discarded the first version and rebuilt the product around that evidence in early 2026. The new design retained context about a firm's work, target accounts, relationships and previous decisions. Research claims linked back to their sources. When the system could not verify a fact, the field remained empty.
Pant wrote a rule for that choice: "An answer that is 80 percent cited and 20 percent blank beats one that is 100 percent complete and 15 percent invented."
The same boundary applied to action. alkemy could research an account and prepare a possible reason to speak, but the relationship owner decided whether to make contact. Through the product's public launch on June 25, 2026, Pant led the thesis, field research, product architecture, specifications, priorities, initial production setup and release. Four engineers in Kathmandu wrote most of the production code, and Aashish Pant subsequently took responsibility for day-to-day development.
When Analysis Gets Cheap
Pant set out the broader argument in a July 12, 2026 essay, "When Analysis Gets Cheap, Relationships Get Expensive".
He described knowledge work as a sequence of gathering information, processing it, analyzing it, acting on the result and accepting responsibility for what follows. AI was becoming faster at the early stages. The final decisions still belonged to people.
"The machine ate the input," Pant wrote. "The human still owns the judgment and the timing."
A 2026 Thomson Reuters Institute survey found that 40 percent of surveyed professionals said their organizations were using generative AI, up from 22 percent the previous year. Pant expects that adoption to lower the cost of research, document analysis, market mapping and first drafts. A person still has to decide which evidence matters, choose when to act and accept the consequences of the recommendation. As firms gain access to similar models, their relationships and accumulated context become harder to reproduce than the analysis itself.
Building Between Nepal and the United States
Most of Pant's market research and client conversations concerned professional firms in the United States, while much of the engineering took place in Kathmandu. Pant translated those conversations into product requirements that engineers could challenge and implement. Joshi and the engineering team determined how the systems would work in production.
Pant frames the arrangement as a test of whether technical teams in Nepal can help define products for international markets. He wants teams to understand the user's problem and its commercial constraints, then make engineering decisions with that context.
Looking Ahead
alkemy is still early, and its commercial outcome remains open. Pant's broader record includes completed client systems, six applied AI products, four that did not find a market and a method that treats adoption as part of product design. His next test is whether the professional firms he studied change how they work after using the product.
