About Me
About Me
Sahil Dua | Product Manager
Sahil Dua
Product Manager
Sahil Dua | Product Manager
I build AI and platform products where technical architecture, user trust, and business outcomes have to work together.
My path into product started in engineering. I spent my early career building ETL pipelines, decision dashboards, and operational systems inside large organizations, the kind of systems where brittle integrations, messy data, and silent failures are not edge cases, but reality.
That foundation shaped how I operate as a Product Manager. I don’t just ask what users want. I ask where the system breaks, what decision needs to improve, and what layer needs to exist for the product to compound.
That pattern has followed me across marketplaces, AI workflows, enterprise automation, and founder-led products.
At Filo, I helped launch the company’s US tutoring marketplace as the founding Product Manager. The easy path was to lift and shift the matching and scheduling logic that worked in India. I rejected that approach because the US market had different liquidity patterns, pricing dynamics, and session intent. We built a US-specific experience instead, scaling the product from 0 to 120K users and $1.5M ARR in six months on a global platform serving 3.3M+ users.
At Supreme Components, I shipped the company’s first GPT-3.5-powered sales automation workflow. The system classified inbound intent, extracted manufacturer part numbers, matched inventory, generated quote drafts, and routed exceptions through human review. It reduced in-stock quote turnaround from 18 hours to under 1 hour and showed me what production AI really requires: not just model capability, but guardrails, workflow design, evaluation, and trust.
I later founded Closphere, an inventory intelligence SaaS for SMBs. I built it the unromantic way, door-to-door across India, talking to business owners and warehouse managers who had no patience for pitch decks and infinite patience for someone who would actually listen. We started as a standalone system, but customer discovery made the real problem clear: SMBs did not want another system of record. They wanted intelligence layered over the tools they already used. We pivoted to an ERP-agnostic plug-in architecture, reduced onboarding from 4+ weeks to 3-7 days, scaled to 63 paying customers across 100+ organizations, and exited through a technology/IP acquisition.
Most recently, at ProductSquads, I worked on AI-native systems where the challenge was not whether the model could extract information, but whether the product could be trusted at production scale. For a document intelligence pipeline failing on long-tail format variance, I audited 200+ failed extractions and pushed the team away from brittle per-format parsers toward a schema-driven retrieval layer. New document variants became configuration changes instead of multi-week engineering projects. The system scaled to 100K+ documents per month at 95%+ accuracy while reducing processing time by 70%.
The lesson across these chapters is consistent: capability is getting cheaper; trust is not. Models hallucinate. Prompts drift. Evaluation is ambiguous. Latency becomes cost. The teams that win with AI will not be the ones that simply ship more features. They will be the ones that make intelligent systems reliable enough for real users, real workflows, and real business decisions.
That is the work I came to the University of Washington to deepen. I’m completing an MS in Information Management with a Product and AI specialization, graduating in August 2026. I’m there to build the product and technical judgment AI systems require, systems that amplify human decision-making rather than replace it.
That same thesis drives NXTai, my graduate project: a decision intelligence platform for product teams. It takes one of the most important but least measurable PM decisions - what to build next and connects fragmented signals from customer feedback, support tickets, CRM data, sales commitments, and internal conversations into a ranked and evidence-backed view of product priorities.
Across everything I’ve built, the work comes back to one question:
How do you make complex systems trustworthy enough that real users will actually rely on them?
I’m looking for senior product roles on teams building AI platforms, workflow automation, developer tools, decision products, or intelligent systems where trust, evaluation, and system design matter as much as user experience.
If that’s the work, let’s talk - sahildua@uw.edu
I build AI and platform products where technical architecture, user trust, and business outcomes have to work together.
My path into product started in engineering. I spent my early career building ETL pipelines, decision dashboards, and operational systems inside large organizations, the kind of systems where brittle integrations, messy data, and silent failures are not edge cases, but reality.
That foundation shaped how I operate as a Product Manager. I don’t just ask what users want. I ask where the system breaks, what decision needs to improve, and what layer needs to exist for the product to compound.
That pattern has followed me across marketplaces, AI workflows, enterprise automation, and founder-led products.
At Filo, I helped launch the company’s US tutoring marketplace as the founding Product Manager. The easy path was to lift and shift the matching and scheduling logic that worked in India. I rejected that approach because the US market had different liquidity patterns, pricing dynamics, and session intent. We built a US-specific experience instead, scaling the product from 0 to 120K users and $1.5M ARR in six months on a global platform serving 3.3M+ users.
At Supreme Components, I shipped the company’s first GPT-3.5-powered sales automation workflow. The system classified inbound intent, extracted manufacturer part numbers, matched inventory, generated quote drafts, and routed exceptions through human review. It reduced in-stock quote turnaround from 18 hours to under 1 hour and showed me what production AI really requires: not just model capability, but guardrails, workflow design, evaluation, and trust.
I later founded Closphere, an inventory intelligence SaaS for SMBs. I built it the unromantic way, door-to-door across India, talking to business owners and warehouse managers who had no patience for pitch decks and infinite patience for someone who would actually listen. We started as a standalone system, but customer discovery made the real problem clear: SMBs did not want another system of record. They wanted intelligence layered over the tools they already used. We pivoted to an ERP-agnostic plug-in architecture, reduced onboarding from 4+ weeks to 3-7 days, scaled to 63 paying customers across 100+ organizations, and exited through a technology/IP acquisition.
Most recently, at ProductSquads, I worked on AI-native systems where the challenge was not whether the model could extract information, but whether the product could be trusted at production scale. For a document intelligence pipeline failing on long-tail format variance, I audited 200+ failed extractions and pushed the team away from brittle per-format parsers toward a schema-driven retrieval layer. New document variants became configuration changes instead of multi-week engineering projects. The system scaled to 100K+ documents per month at 95%+ accuracy while reducing processing time by 70%.
The lesson across these chapters is consistent: capability is getting cheaper; trust is not. Models hallucinate. Prompts drift. Evaluation is ambiguous. Latency becomes cost. The teams that win with AI will not be the ones that simply ship more features. They will be the ones that make intelligent systems reliable enough for real users, real workflows, and real business decisions.
That is the work I came to the University of Washington to deepen. I’m completing an MS in Information Management with a Product and AI specialization, graduating in August 2026. I’m there to build the product and technical judgment AI systems require, systems that amplify human decision-making rather than replace it.
That same thesis drives NXTai, my graduate project: a decision intelligence platform for product teams. It takes one of the most important but least measurable PM decisions - what to build next and connects fragmented signals from customer feedback, support tickets, CRM data, sales commitments, and internal conversations into a ranked and evidence-backed view of product priorities.
Across everything I’ve built, the work comes back to one question:
How do you make complex systems trustworthy enough that real users will actually rely on them?
I’m looking for senior product roles on teams building AI platforms, workflow automation, developer tools, decision products, or intelligent systems where trust, evaluation, and system design matter as much as user experience.
If that’s the work, let’s talk - sahildua@uw.edu
About Me
I build AI and platform products where technical architecture, user trust, and business outcomes have to work together.
My path into product started in engineering. I spent my early career building ETL pipelines, decision dashboards, and operational systems inside large organizations, the kind of systems where brittle integrations, messy data, and silent failures are not edge cases, but reality.
That foundation shaped how I operate as a Product Manager. I don’t just ask what users want. I ask where the system breaks, what decision needs to improve, and what layer needs to exist for the product to compound.
That pattern has followed me across marketplaces, AI workflows, enterprise automation, and founder-led products.
At Filo, I helped launch the company’s US tutoring marketplace as the founding Product Manager. The easy path was to lift and shift the matching and scheduling logic that worked in India. I rejected that approach because the US market had different liquidity patterns, pricing dynamics, and session intent. We built a US-specific experience instead, scaling the product from 0 to 120K users and $1.5M ARR in six months on a global platform serving 3.3M+ users.
At Supreme Components, I shipped the company’s first GPT-3.5-powered sales automation workflow. The system classified inbound intent, extracted manufacturer part numbers, matched inventory, generated quote drafts, and routed exceptions through human review. It reduced in-stock quote turnaround from 18 hours to under 1 hour and showed me what production AI really requires: not just model capability, but guardrails, workflow design, evaluation, and trust.
I later founded Closphere, an inventory intelligence SaaS for SMBs. I built it the unromantic way, door-to-door across India, talking to business owners and warehouse managers who had no patience for pitch decks and infinite patience for someone who would actually listen. We started as a standalone system, but customer discovery made the real problem clear: SMBs did not want another system of record. They wanted intelligence layered over the tools they already used. We pivoted to an ERP-agnostic plug-in architecture, reduced onboarding from 4+ weeks to 3-7 days, scaled to 63 paying customers across 100+ organizations, and exited through a technology/IP acquisition.
Most recently, at ProductSquads, I worked on AI-native systems where the challenge was not whether the model could extract information, but whether the product could be trusted at production scale. For a document intelligence pipeline failing on long-tail format variance, I audited 200+ failed extractions and pushed the team away from brittle per-format parsers toward a schema-driven retrieval layer. New document variants became configuration changes instead of multi-week engineering projects. The system scaled to 100K+ documents per month at 95%+ accuracy while reducing processing time by 70%.
The lesson across these chapters is consistent: capability is getting cheaper; trust is not. Models hallucinate. Prompts drift. Evaluation is ambiguous. Latency becomes cost. The teams that win with AI will not be the ones that simply ship more features. They will be the ones that make intelligent systems reliable enough for real users, real workflows, and real business decisions.
That is the work I came to the University of Washington to deepen. I’m completing an MS in Information Management with a Product and AI specialization, graduating in August 2026. I’m there to build the product and technical judgment AI systems require, systems that amplify human decision-making rather than replace it.
That same thesis drives NXTai, my graduate project: a decision intelligence platform for product teams. It takes one of the most important but least measurable PM decisions - what to build next and connects fragmented signals from customer feedback, support tickets, CRM data, sales commitments, and internal conversations into a ranked and evidence-backed view of product priorities.
Across everything I’ve built, the work comes back to one question:
How do you make complex systems trustworthy enough that real users will actually rely on them?
I’m looking for senior product roles on teams building AI platforms, workflow automation, developer tools, decision products, or intelligent systems where trust, evaluation, and system design matter as much as user experience.
If that’s the work, let’s talk - sahildua@uw.edu