Keval HirparaAI Product Manager ↗
Rec · cam_01 · ego viewOpen to AI Product Manager roles

person · keval_hirpara · 0.99KevalHirpara.

role · 0.97AI Product Manager

[caption] I make sure the data robots learn from is actually usable. AI features earn people’s trust. every launch has a baseline and a metric. teams ship what users actually need.

Now

Open to AI product roles

Last deployment

Founder’s Office, Human Archive (YC W26)

B.Tech in Electrical Engineering, NIT Surat

▸ Play the case studies▸ Play the case studiesOpen a channel
Scroll to scrub the recording

scene 01 · About

Product thinking, with a builder’s hands.

I’m Keval, an AI product manager in Bengaluru. I work on the problems between AI models and the real world — how training data gets collected, how quality is measured, and how AI features earn users’ trust.

In the Founder’s Office at Human Archive (YC W26), a robotics and embodied-AI data startup, I tracked the end-to-end data pipeline — from on-robot collection to AWS storage — where I learned that the hardest product questions in AI are often about data: what counts as usable, how failures are caught, and what they cost. My case studies go deep on exactly that.

I studied Electrical Engineering at NIT Surat and came up as a builder — a software development internship, full-stack apps, AI agent pipelines, and 1,100+ competitive programming problems. It means I can go deep with engineers on feasibility and trade-offs, then zoom back out to the user and the metric.

Education

B.Tech in Electrical Engineering

NIT Surat

2022 – 2026

Achievements

  • Winner, FINFIESTA 2023 — organized by CEV, NIT Surat
  • 1,100+ algorithmic problems solved across LeetCode and Codeforces; peak LeetCode rating 1642
  • Participant, Medecro HealthHack 2024 — organized by Medecro.AI

Experience

  1. Feb 2026 – Aug 2026

    Founder’s Office

    Human Archive (YC W26)

    Robotics / embodied-AI data startup · Gurugram, India

    • Tracked and analyzed the end-to-end data pipeline — on-robot collection, processing, AWS-based storage — surfacing bottlenecks, delays and data-sync gaps across cross-functional teams.
    • Coordinated deployment, data collection, QA and offloading teams to keep operational workflows moving and unblock stalled handoffs.
    • Monitored pipeline stability and throughput, feeding operational visibility back into deployment and infrastructure priorities.
    • Ideated and built an internal Deployment & Inventory Tracker giving the team one view of device status and pending work.

  2. Jun 2025 – Jul 2025

    Software Development Intern

    Dvij InfoTech

    Surat, India

    • Built full-stack features on the MERN stack (MongoDB, Express.js, Next.js, React.js, Node.js), translating functional requirements into working application components.
    • Improved front-end responsiveness and refactored components to reduce technical debt and improve long-term maintainability.

  3. Aug 2024 – Mar 2025

    Convenor, Co-Curricular Affairs Council

    NIT Surat

    Leadership

    • Led 400+ volunteers to run 30+ events for 5,000+ participants at MindBend, Gujarat’s largest techno-managerial fest — owning planning, timelines and resource allocation.
    • Designed and ran a multi-channel outreach strategy across 50+ colleges, driving a 30% year-over-year increase in participation.

Product

  • Problem framing
  • Requirements & process mapping
  • North-star & metric trees
  • Prioritization (RICE, MoSCoW)
  • Experiment design
  • Stakeholder communication

AI & Data

  • Embodied AI & robotics data
  • Data pipelines & quality
  • Root cause analysis
  • LLM agents & orchestration
  • Operational reporting
  • Unit economics

Technical

  • SQL
  • Python (Pandas, NumPy)
  • JavaScript & TypeScript
  • React & Next.js
  • Node.js & MongoDB
  • C++ & DSA

Tools

  • Jira
  • Notion
  • Excel & Power BI
  • Git & GitHub
  • Google Colab
  • Claude Code

QA review: what I keep.

4 frames reviewed · 4 rejected · 4 usable

  1. Frame 001

    ✕ Rejectedroadmaps full of features nobody measured

    ✓ Usablea baseline, a target metric and a decision I can defend.

  2. Frame 002

    ✕ RejectedAI demos that only work on the happy path

    ✓ Usableproducts that fail gracefully and earn trust on the bad days.

  3. Frame 003

    ✕ Rejectedblaming the model first

    ✓ Usablechecking the data before anything else.

  4. Frame 004

    ✕ Rejectedspecs thrown over the wall

    ✓ UsablePMs who can read the code and sit with the people using it.

If you’re building AI that has to work outside the demo, let’s talk. Open a channel →

scene 02 · Case Studies

Product case studies

Problem framing, metrics, prioritization and rollout — the full reasoning, start to finish.

scene 03 · Projects

Things I’ve built

20 projects · 2025—2026
All repositories on GitHub ↗

scene 04 · Playground

Small experiments, just for fun

Open the full playground →

01Dot Field

Canvas

A grid of points that gets pushed around by your cursor. Canvas + spring physics.

02Sort Visualizer

Algorithms

Bubble, insertion, and selection sort, step by step. A nod to my DSA roots.

03BFS Pathfinder

Graphs
A
B
Click or drag to draw walls.

Draw walls, then watch breadth-first search flood the grid and find the shortest path.

04Kinetic Type

Typography

Type your name, then move your cursor (or drag a finger) across the letters to watch them swell and lean.

scene 05 · Contact

Let’s build something together.

Email

kevalhirpara2003@gmail.com

Location

Bengaluru, India

Social

I’m reaching out about