Work
My work has gradually expanded from solving high-context technical problems to building the systems around how engineering teams operate. Today that includes platform and integration architecture, enterprise systems, developer experience, reliability, AI-assisted engineering, execution mechanisms, and people development. The platform north star is a golden path: reliability and performance first, then shared leverage, then enterprise and AI readiness. I still stay close to architecture — most of the underlying systems are ones I have designed or reviewed — and I measure impact by what continues to work when I am not directly involved.
SpotDraft. Leading engineering across Integrations and Platform. I stay close to architecture and code while managing 15+ people: most of the underlying platform and integration systems are ones I have designed or reviewed, and about 90% of the engineering organization comes to me for architectural decisions. The 2026 platform charter is reliability first, then shared leverage, then enterprise and AI readiness — including multi-workspace identity and an MCP tool registry.
Questionnaire foundations first, then workflow systems, then integrations and Legal Intake, and now leadership across integrations, platform systems, developer experience, AI-assisted engineering, and people development.
Leading engineering across Integrations and Platform, with work spanning platform architecture, developer experience, reliability, AI-assisted engineering, and people development. I stay close to architecture while managing 15+ people: most of the underlying systems are ones I have designed or reviewed, and about 90% of the engineering organization comes to me for architectural decisions.
Moved from core product systems into integrations, helped shape automation-oriented design for CRM surfaces, and later drove Legal Intake from internal hackathon concept toward a shipped product direction.
Built the self-serve questionnaire experience for template workflows, extended the same foundation to third-party paper without destructive rewrites, and then helped build the current Workflow Manager on top of that broader workflow foundation.
Worked on internal fulfillment systems in Amazon's European Fulfillment Network, spanning requirements engineering, feature design, data pipelines, marketplace-launch automation, and high-scale inventory systems.
Built practical systems across IoT, full-stack product work, systems programming, and applied computing, including an IoT plant irrigation project and other open-source experiments.
Worked with Chapel Language as part of Google Summer of Code 2019 on the Unit Testing Framework for Chapel, which deepened my grounding in systems programming, testing infrastructure, and developer tooling.
I led SpotDraft's agentic software development lifecycle. The job was to take AI out of individual prompting and put it into the actual engineering loop: requirements and acceptance criteria, planning, implementation, review, tests, and merge-ready PRs. I stood up the working group, defined that loop, and made backend and frontend codebases agent-ready so conventions and review standards did not have to be re-taught on every ticket. We built an orchestrated spec-to-PR workflow, then drove adoption across teams instead of leaving it as a side-channel POC. The remaining work is the same as any other engineering system: speed, orchestration quality, and whether the output is actually review-ready.
SpotDraft MCP is the agent gateway for the product. Sidebar, automations, Legal Intake, and external coding agents consume the same server. Adding a tool used to mean wiring it through every consumer by hand. I am designing the architecture so each tool is tagged by capability, area, and maturity, and an audience is a filter on those tags — not a hardcoded list. Preview stays out of production agents until it is evaluated. That is how a pod registers a tool without breaking everyone else, which is the same golden-path bet as the rest of the platform.
Developer productivity often breaks down because engineers repeatedly pay the same coordination or tooling cost. I try to identify those patterns and turn them into better defaults. I'm particularly interested in the point where developer productivity stops being an individual habit and becomes an engineering system.
Good engineering organizations need space for ideas that do not begin on the quarterly roadmap. For the last two years, I've helped organize and host SpotDraft's company hackathons, creating space for engineers and cross-functional teams to experiment, collaborate outside their normal boundaries, and turn ideas into working prototypes. I see hackathons as more than events. They are a mechanism for making experimentation culturally safe and for discovering ideas that normal roadmap processes might never surface.
An engineering manager's job is not to make every important decision. It is to develop engineers who can make increasingly important ones themselves. Managing 15+ people is a judgment-transfer problem. A harder ticket with the same decision-maker behind it is still a ticket. Someone owns the work when they choose the approach, explain the tradeoff, and live with the result — including on-call and the stakeholder conversation.
They finish a well-bounded piece of work end to end: implementation, tests, review, follow-through. The point is being able to explain the tradeoff, not waiting for the next instruction.
They hold the spec, the sequence, and the stakeholder conversation. Continuous feedback and an explicit growth plan sit underneath, but the assignment is the mechanism.
They can argue the contract, the failure mode, and the edge case — not only implement a design I already chose. Architecture reviews and production responsibility are how that judgment forms.
They set the bar for others: reviews, on-call, mentoring, and the call that used to default to me. After the project ships, can they own the next similar problem? If not, we delivered. We did not mentor.
Promotion cases from explicit growth plans, stretch ownership, and evidence against next-level expectations. The packet is not the work. The work is a decision space the person can occupy without me covering fire.
Native integrations, on-call, specs, architecture reviews, roadmap discussions, and operational systems. Those are easy to hoard as a manager who is still close to the work. If the important call still routes through me by default, they do not own it yet.
I still review architecture, contracts, risks, and edge cases. Implementation quality should move to peer and tech-lead review. About 90% of engineering still comes for the hard call. The people job is to shrink that load without dropping the bar.
The full argument: widen the decision space, review at the right altitude, and do not take the pen back after a miss.
Worked on moving integrations away from bespoke customer-specific implementations toward reusable primitives and a generic trigger-condition-action automation model. One example was choosing a reusable Reverse Sync architecture rather than optimizing only for the fastest Salesforce implementation, so later CRM, Slack, and access-control automations could reuse the same underlying system. Salesforce and HubSpot native integrations have each had more than $1M in revenue impact.
The platform job is to take reliability out of heroics: stop using the primary database as a queue, remove single points of failure in routing, and make the product feel live instead of stale. Dashboards, ownership, incident follow-through, and on-call are how that operating system stays inspectable. The most important shift was from treating reliability as a sequence of incidents to treating it as an engineering system.
Auth, access, tasking, automation, search, bulk operations, and shared UI should be foundations product teams consume, not puzzles they re-solve. I am pushing the org toward that golden path: explicit tenant ownership, modular frontend platforms, and shared services so adding an audit log, a search index, or an AI tool takes minutes instead of a custom project.
Led SpotDraft's agentic SDLC: agent-ready backend and frontend codebases, an orchestrated spec-to-PR workflow, and adoption across teams. The goal was a development system that produces review-ready work, not more AI-tool usage. The same pillar includes MCP infrastructure so product teams can register agent tools on a shared gateway instead of each inventing one.
Helped build a more self-service developer experience through improvements to SpotDraft's public APIs, API documentation, Public API UI, and Developer Portal. That is how external builders consume the platform. Internally, the larger bet is the golden path above — not a portal as the product.
Legal Intake was an important lesson in treating execution itself as something that can be engineered. I introduced spec-first planning, phased milestones, explicit ownership, Jira-based execution, QA ownership, visible risk tracking, and reusable planning templates. The value was not only shipping the project. It was creating a planning model that could be reused on later initiatives. Estimated revenue impact by the end of this financial year is more than $1M.
Building self-service systems, tooling, automation, documentation, APIs, developer infrastructure, and engineering workflows that reduce repeated coordination cost.
Led promotion of 7+ engineers through explicit growth plans, stretch ownership, technical coaching, continuous feedback, and evidence-based promotion readiness.
Recognized with SpotDraft's Excellence Award for going above and beyond as an engineering manager, with specific recognition for impact in leveling up the PED function and the product.
Open a project for challenge, approach, and outcome. Recognition stays on the row when it is part of the public record.
A product-systems arc that started with self-serve questionnaire setup for template workflows, was extended to third-party paper on the same foundation, and then evolved into the current Workflow Manager.
A configurable reverse-sync system for pushing contract and workflow data back into Salesforce, shaped to be extensible across triggers, conditions, object mappings, and future automation use cases.
Salesforce native integrations: more than $1M in revenue impact.
An internal 2024 hackathon project focused on structured legal-request intake, built with Sanket Mishra and Punit Batra, awarded third prize, and later shaped into a shipped product surface led by me.
Third prize, 2024 internal hackathon. Estimated revenue impact by end of this financial year: more than $1M.
Native integration and platform work spanning CRM, collaboration, and adjacent systems, with emphasis on reusable primitives, metadata-aware design, and integration surfaces that stay coherent as the catalog expands.
HubSpot native integrations: more than $1M in revenue impact.
ClickWrap captures recorded acceptance of terms in-product — a checkbox, link, or agreement viewer on a website, app, or signup flow — without sending the agreement through a full negotiation and e-sign ceremony. We explored that product shape as a 2023 hackathon with Sanket Mishra and Punit Batra; another team later shipped it. The shipped product has had more than $1M in revenue impact.
Third prize, 2023 internal hackathon. Shipped-product revenue impact: more than $1M.
An Amazon EFN chapter focused on internal fulfillment systems, requirements engineering, data pipelines, marketplace-launch automation, and high-scale inventory services where correctness and latency both mattered.
My Google Summer of Code 2019 project with Chapel Language focused on the Unit Testing Framework for Chapel, which gave me an early grounding in systems-level tooling and test infrastructure.
Krishna Dey