Skip to content

Open to Software Engineering roles

Utsav Kumar · Software Engineer (AI & Full-Stack) in Vancouver, Canada

I build AI agents that hold up in production.

Software Engineer with 10+ years building full-stack products, now with AI at the core. I work on the hard part of production AI: latency, cost, context, grounding, and reliability, so people actually trust what the model says.

  • LLMs
  • AI Agents
  • Multi-agent orchestration
  • RAG
  • Embeddings
  • Vector search
  • Hybrid search & re-ranking
  • Tool calling
  • MCP
  • Structured outputs
  • Context engineering
  • Conversation memory
  • Grounding & citations
  • Token optimization
  • Inference latency
  • Semantic caching
  • Cost optimization
  • Streaming (SSE)
  • Prompt design
  • Workflow routing
  • Guardrails
  • Human-in-the-loop
  • Evals & rubrics
  • LLM observability
  • RLHF training data
See my workLinkedIn

Previously at

Based in Vancouver, BC

  • Highspot2023 – 2026
  • Mercor2025 – 2026
  • Splunk2018 – 2023
Years building full-stack products
10+
p95 time to first tokendown from 8s
0.8s
Agent task successup from 78% · 500 eval cases
91%
Agent workflows completed per month
12k
Lower AI compute costwith result caching
~80%
Faster deal preparation15 min → 5 min
3×
Weekly adoption among eligible usersup from 35% in 3 months
60%
ARR enabled by a critical release
$10M

AI projects

AI products I've designed, built, and shipped.

Highspot

Deal, Rep & Manager Agents

A family of AI assistants for sales teams: deal-level answers, portfolio-wide insights, and coaching prep for leaders, built on one shared agent platform.

  • AI-driven workflow selection, business tools, and conversation memory
  • Live token streaming over SSE with human approval for sensitive actions
  • Permissions, tenant controls, and feature flags on every data access
weekly active reps across 10 customer orgs
150
agent workflows per month
12k
of first-time users return within 7 days
65%

AI & stack

  • AI Agents
  • LLMs
  • PydanticAI
  • Tool calling
  • Workflow routing
  • Conversation memory
  • RAG
  • Structured outputs
  • Result caching
  • SSE streaming
  • Human-in-the-loop
  • Evals
  • Python
  • FastAPI
  • React
  • TypeScript

Highspot

AI Sales Coaching & Role-Play

Real-time coaching, feedback, and practice conversations that help sales reps improve, grounded in trusted content and streamed as they are generated.

  • AI Sales Coaching with personalized, real-time feedback for reps
  • AI Role-Play practice conversations streamed live over SSE
  • Turned AI concepts into intuitive experiences with PMs and designers
role-play sessions completed per month
2,000+
time to first streamed feedback token
<1s
average rep rating of AI feedback
4.5/5

AI & stack

  • OpenAI
  • LLMs
  • RAG
  • Grounding
  • Prompt design
  • Context engineering
  • SSE streaming
  • React
  • TypeScript
  • Python

Mercor

AI Model Evaluation

Expert Project Lead on AI training and evaluation, measuring how well models actually perform on complex software engineering and coding tasks.

  • Designed prompts, grading rubrics, and evaluation criteria for coding tasks
  • Graded model responses for correctness, missing requirements, and reasoning errors
  • Led reviewer calibration sessions and quality-checked expert submissions
model-response evaluations completed
1,200+
reviewer agreement after calibration
95%
engineering experts calibrated and reviewed
25+

AI & stack

  • Evals
  • Rubric design
  • Prompt design
  • Model grading
  • RLHF training data
  • Reviewer calibration
  • Code review

About

Taking AI from a great demo to a dependable product. That's where I do my best work.

I'm a Software Engineer based in Vancouver. My career in Canada began with a Master's in Computer Science, and I've been building products here ever since.

At Splunk, I worked across Splunk Enterprise, Data Stream Processor, and the Edge Processor UI, building the tools customers use to collect, transform, and route their data.

At Highspot, I spent nearly three years as a project lead, taking projects end to end, from the first idea all the way to the finish line.

Today my focus is AI: building AI agents and LLM-powered features in Python and TypeScript that people can trust every day. I also use AI almost every day in my own work, with Cursor, Claude Code, and Codex as part of how I plan, write, and review code.

Outside work

  • Squash
  • Travel
  • Pool

Education

Grounded in computing, sharpened in production.

Experience

A decade across startups and large enterprises.

  1. Jun 2023 – Aug 2026

    Vancouver, BC

    HighspotSenior Software Development Engineer

    • Built and shipped the AI-powered Deal Agent, Rep Agent, and Manager Agent with React, TypeScript, Python, FastAPI, PydanticAI, and LLMs, used by 150 weekly active reps across 10 customer organizations and completing 12,000 workflows a month.
    • Improved agent task success from 78% to 91% across 500 evaluation cases, and reduced p95 time to first token from 8 seconds to 0.8 seconds.
    • Used React Query and caching to cut unnecessary AI requests, reducing AI compute costs by 80% and improving UI response times by 90%.
    Show 10 more
    • Reduced deal preparation time from 15 minutes to 5 minutes.
    • Grew weekly adoption among eligible users from 35% to 60% over 3 months, with 65% of first-time users returning within 7 days.
    • Connected agents to internal services through async FastAPI APIs with validation, permissions, tenant controls, and feature flags.
    • Deal Agent: an assistant for individual opportunities using AI-driven workflow selection, business tools, conversation memory, and live streaming.
    • Rep Agent: analyzes many deals and accounts, retrieves trusted business data, and gives sellers prioritized insights and coaching.
    • Manager Agent: built on the shared Rep Agent platform to help leaders review pipelines and spot deals and reps needing attention.
    • Added monitoring and human approval for actions requiring confirmation, keeping customer data access controlled.
    • Instrumented AI experiences with Amplitude to surface friction and guide product priorities.
    • Partnered with PMs and designers on reusable components, reducing development time by 20%.
    • Contributed to architecture, roadmap planning, estimation, code reviews, and mentoring junior engineers.

  2. Feb 2025 – Jul 2026

    Remote · Contract

    MercorSoftware Engineering Expert · Expert Project Lead

    • Completed 1,200+ evaluations of model-generated responses on complex software engineering and coding tasks against rubrics and reference answers.
    • Designed prompts, grading rubrics, and evaluation criteria that were clear, verifiable, and consistent across reviewers.
    • Led calibration sessions with engineering experts and quality-checked their submissions before acceptance.

  3. Feb 2019 – Feb 2023

    Vancouver, BC

    SplunkSoftware Engineer

    • Led front-end development for the Edge Processor team, building real-time data processing and transformation experiences.
    • Parallelized GitLab CI/CD pipelines, reducing job runtimes by 80%.
    • Helped the Data Manager team ship a critical release on time, contributing to $10M in ARR, and earned a SPOT award.
    Show 6 more
    • Built features for Data Stream Processor so customers could visually configure, transform, and route data streams.
    • Improved the Pipeline Canvas Builder with reusable components, cutting pipeline setup time by 30%.
    • Sped up Edge Processor pages by optimizing asset delivery and removing unnecessary API calls.
    • Reached 95% coverage with Cypress and Jest, and introduced Mock Service Worker for backend-independent development.
    • Documented reusable components in Storybook with Splunk's UI Design System.
    • Led frontend knowledge sharing through tech talks, code reviews, and team syncs.

  4. Sep 2018 – Dec 2018

    Vancouver, BC

    SplunkAgile Full Stack Developer Co-op

    • Built frontend features and reusable Storybook components for DB Connect, Splunk's enterprise data source integration.

  5. Jun 2015 – Aug 2017

    Gandhinagar, India

    Cybage Software Pvt LtdSoftware Engineer

    • Built features for Turn Campaign Suite, a platform for managing digital advertising campaigns, alongside onshore teams.
    • Reached 90% unit test coverage with Jasmine and expanded Selenium coverage for critical workflows.
    • Ran root cause analyses for production issues and organized team retrospectives.

  6. Dec 2013 – Jun 2015

    Ahmedabad, India

    WeblineindiaJunior Web Developer · Web Developer

    • Built responsive client websites and custom WordPress themes from PSD designs using HTML, CSS, JavaScript, jQuery, LESS, SASS, and AngularJS.

Toolkit

From the React frontend to the agent layer.

AI layer

  • OpenAI
  • LLMs
  • AI Agents
  • PydanticAI
  • LangChain
  • RAG
  • Embeddings
  • Vector search
  • Pinecone
  • pgvector
  • MCP
  • Tool calling
  • Structured outputs
  • Prompt engineering
  • Context engineering
  • Conversation memory
  • SSE streaming
  • Guardrails
  • Evals

Frontend

  • React
  • TypeScript
  • JavaScript
  • Next.js
  • React Query
  • Redux
  • HTML
  • CSS
  • Sass
  • Tailwind CSS
  • Storybook
  • Design systems
  • Responsive design
  • Web performance

Backend

  • Python
  • FastAPI
  • Pydantic
  • Node.js
  • Ruby on Rails
  • REST APIs
  • Async APIs
  • PostgreSQL
  • Caching
  • Auth & permissions
  • Multi-tenant systems
  • Feature flags

Quality & delivery

  • Cypress
  • Jest
  • Jasmine
  • Selenium
  • MSW
  • GitLab CI/CD
  • Code review
  • Root cause analysis
  • Amplitude
  • Agile
  • Technical planning
  • Mentoring

AI tools I use daily

  • Cursor
  • Claude Code
  • Codex
  • ChatGPT
  • Claude
  • AI-assisted code review
  • Agentic coding

Contact

Building something with AI? Let's talk.

I'm open to software engineering roles on AI products and full-stack teams. Send a message and it comes straight to my inbox.