Harshit Sharma
AI Product Manager
Product manager building at the frontier. I ship real products, work across the AI stack, and think obsessively about evals, reliability, and what makes AI systems trustworthy.
Selected Work, 6 Projects
Eval Studio
Which prompt, which model, at what cost?
The Problem
AI teams pick prompts and models using spreadsheets, gut feel, or generic benchmarks that have nothing to do with their actual product.
The Solution
Browser-based LLM eval tool. Test prompts and models on your own data with multi-model judge council, cost tracking, and ranked results.
claude-code-bridge
Context compounds. It never resets.
The Problem
Every time you switch from Claude to Cursor, context is lost. You re-explain decisions, repeat constraints, and rebuild mental state from scratch.
The Solution
A local MCP server that auto-syncs strategy decisions from Claude to your project files. When you open Cursor, Claude Code already knows where things stand.
Explainable AI Coding Assistant
Trust Through Transparency
The Problem
Developers waste hours verifying AI suggestions because tools optimize for speed, not trust, creating a “trust tax” that blocks adoption.
The Solution
Calibrated confidence scores + explainable reasoning. Show developers WHY a suggestion works and WHEN to be skeptical.
Job Market Pulse
The labor market, visualized
The Problem
Job market data is scattered across government sites, ugly portals, and paywalled reports. Nobody has a clean, real-time view of what is actually happening.
The Solution
A real-time dashboard pulling from FRED API, DOL, and USCIS. Labor market trends, H-1B sponsor rankings, PM salary ranges, and approval rates in one view.
Dear Her
Some feelings are too big for a text message
The Problem
Most people feel things deeply about the women in their lives but never find the words to say it. The gap is not feeling. It is articulation.
The Solution
Write three honest prompts about her. Claude transforms them into a beautiful animated letter she can open from any link, no login required.
PM Salary Ace
Practice Like the Job Depends On It
The Problem
PM candidates don't know what skill level they're actually at. Generic prep doesn't map to real compensation gaps.
The Solution
336 questions across 5 salary tiers. Practice what a $350K Staff+ PM actually needs, not generic frameworks.
Background
Building AI Products That Scale Reliably
I started in electrical engineering, designing power systems where failure meant explosions, not just bad user reviews. That taught me to think through failure modes obsessively because in safety-critical systems, “it works in the demo” isn’t good enough.
For five years I shipped products at scale: complex enterprise software, data platforms, and revenue tooling that supported a $150M+ sales pipeline. I learned how to work with large cross-functional teams, drive adoption across organizations, and own outcomes end to end.
At Berkeley (M. Eng.), I’ve sharpened both edges: technical depth in ML, deep learning, and RL, combined with hands-on AI evals productization. I build products daily, not just study them.
I’m a daily AI power user: I code with Claude, build agents for automation, and live in the products I want to help build.
My focus: bridging the gap between AI capability and reliable deployment at scale because the companies that win won’t just have the most capable models, they’ll have the most trustworthy ones.
WHERE I STARTED
Non-traditional: safety-critical systems to enterprise SaaS. I’ve worked where failure = lawsuits and casualties, not just bugs.
WHERE I AM
Berkeley MEng. Deep in ML/RL/AI evals. Daily power user across frontier models. Building real products, shipping them, iterating.
WHERE I’M GOING
Making AI systems reliable enough to trust in production. Where models are powerful but reliability is everything. Hard problems, high stakes.
Connect
Let's Connect!
Interested in AI product strategy? Building something cool? Just want to chat about agents and reliability? I'm easy to reach.
Currently: Getting back into chess, dominating the table tennis table, and planning my next trip to Japan.
Email me
harshit@harshit.ai
Connect with me on LinkedIn
/in/harryleads
Check out my code
/harshitleads
Grab my resume
HarshitS_Resume.pdf
If it's easier, pick a time directly
Quick intro call · 15 min · we can go over if we're on a roll :)