Mansi - AI Product Manager Portfolio

Unscript AI

Associate PM • 0-to-1 AI Video Generation

The Problem

Video creators need multiple subscriptions to generate AI videos: Heygen for avatars, Synthesia for scene composition, plus manual editing. No single tool does end-to-end video generation. The result: expensive, fragmented, and requires in-house editing teams.

What I Owned

Full product lifecycle: market research, user discovery, PRD, backlog management, sprint planning, wireframes, flows, acceptance criteria, stakeholder management

Led 5-person team: 2 backend engineers, 2 AI engineers, 1 UX designer (reported to founder)

The Key Decision: Avatar-First

Hypothesis: If avatar quality is poor, users won't proceed to scene generation or video creation. Avatar is the bottleneck, not downstream features.

I validated this with user research and presented the reasoning to the dev team. The decision was to prioritize avatar creation first, even though it delayed other features. This meant saying no to rapid feature expansion and focusing on quality gates.

What We Shipped

Avatar creation flow with quality fine-tuning (product images from 10+ angles)

Scene generation workflow (layout, positioning, branding)

Image-to-video pipeline with product integration

Final video stitching (end-to-end flow)

Planned but not shipped: Script, voiceover, transcription

Proof Points

25% reduction in time-to-complete features through workflow standardization and sprint streamlining (before: ad-hoc, after: structured)

Led cross-functional team without direct reports (influence-based leadership)

Mastered AI domain rapidly (new to me; learned embeddings, fine-tuning, API integration)

What I Learned

Not every product ships. Despite solid product thinking and team execution, the founder pivoted before launch. This taught me that not every decision is yours to make, and context matters more than the quality of your work sometimes. What remains: evidence that I can think through complex 0-to-1 problems, reason about prioritization, and execute with small teams.

AI Interview Buddy

Founder • Voice-First Mock Interview Platform

Why I Built It

I've been there: transitioning into PM, not knowing where to start, how to prepare, or how to reduce interview anxiety. Generic AI mock interview tools exist, but none are designed for PM career changers. I saw the gap from personal experience.

Validation (10-15 Users)

Common pain: "I don't know where to start" + "How do I actually prepare for product thinking?"

Existing solution gap: Mock interview tools are generic (can't teach PM-specific thinking)

Anxiety driver: Real interviews feel unpredictable; they need practice with natural conversation flow, not rigid Q&A

Validated need: "I need feedback on my thinking, not just my answers"

Hypothesis

Natural, flowing conversation + adaptive questions that respond to user answers = reduced anxiety + better prep + confidence for real interviews.

Current State (MVP)

Built prototype with Lovable (full interview journey mapped)

What works: Interface, flow, user journey feel natural

What doesn't: Transcription is broken (API misses words → feedback loop breaks); AI voice is robotic

The Blocker (And My Thinking)

Problem: Lovable free tier can't modify APIs. Current transcription API misses words, breaking feedback. I can't user test until this is fixed (no feedback = broken hypothesis).

Constraint: Jobless, no funds for paid APIs or Lovable pro.

Strategic response: This isn't a reason to stop. It's a reason to sequence differently. Fix transcription first, then voice quality, then LLM adaptation. Don't test with users until the core experience works.

Roadmap (What's Next)

Phase 1: Integrate Whisper API (real-time transcription) — unblocks feedback loop

Phase 2: ElevenLabs for natural voice (replaces robotic voice) — improves UX

Phase 3: LLM adaptation (questions respond to user answers) — core hypothesis

Then: Ship and test with actual PM career changers

What This Shows

Problem identification from personal experience (not guessing)

Validation through user interviews (10-15 data points)

Prototyping under constraint (Lovable free tier; creative approach)

Strategic roadmapping (sequence by technical dependency, not by "cool factor")

Technical awareness (knows Whisper API, ElevenLabs, LLM adaptation; understands blocker vs. preference)