A desk with wireframe sketches, sticky notes and a phone

I pick real problems and build solutions using AI.

Product manager with 8+ years across consulting and product, now taking AI products from problem statement to production.

What I'm building

Six products, one pattern: each started as a problem space worth exploring, then became a working attempt to solve it with AI.

Watch the six products in just over two minutes. Product screens are the real apps; cinematic shots, music and voice were generated with Google Veo, Lyria and Chirp.
A thick prospectus flagged with paper tabs beside a phone showing a message with a PDF

IPO Analyst Agents

Runs daily

Multi-agent IPO research

An IPO prospectus runs to hundreds of pages, and most retail investors decide on a headline and a grey-market rumour. Here 12+ AI agents, orchestrated by the LangGraph framework, read the prospectus of every Indian mainboard IPO open that day: eight specialist agents each analyse one specific section in parallel, others handle peer valuation, market sentiment and synthesis, and an advisor agent makes a BUY / NO BUY call. A plain-English summary and a full PDF report land on Telegram, followed by daily GMP and subscription updates until the IPO closes.

  • Models read, Python decides the numbers: every figure carries its prospectus page and a verbatim quote, and a verifier sends failed fields back to the agent (max 2 re-asks), never guessing
  • Every summary is scored by a rulebook and an LLM judge before it is sent, with one revision allowed
  • Hard $5 cost cap per IPO on a metered ledger, model routing across four tiers, and 550+ offline tests that run the real graph

Python · LangGraph · LangChain · OpenAI · Groq · Telegram · WeasyPrint

View on GitHub
Hands arranging paper shapes beside a tablet

Ozzy

Prototype → production

AI literacy for kids

If AI is going to automate the work, people need to know how to direct it. Ozzy teaches kids that skill, the craft of prompting, through play, built on one insight: AI's speed is the earned reward of a good prompt, not a freebie. A vague prompt visibly fails; a precise one succeeds and produces something the child keeps, across a connected Learn → Make → Collect arc.

  • Teaches the five skills of directing AI: specification, context, constraints, examples, decomposition
  • Create-first chapters, where every lesson ends by making a real artifact
  • Stars for precision, boss levels, and a gallery of everything made

React · TypeScript · Real-AI generation

View live
A phone photographing a plastic bottle above four waste bins

Pic N Choose

Live MVP

AI waste segregation

India's SWM-2026 rules make households sort waste into four streams (wet, dry, sanitary, special-care), but most people genuinely don't know which bin an item belongs in, so recyclables get contaminated and everything ends up mixed. Pic N Choose removes the guesswork: photograph or upload any item, AI identifies it, and a rules engine maps it to the correct stream with the reason and the right handling, so the decision that used to take a Google search takes a second.

  • AI photo classification paired with a deterministic rules engine, so the taxonomy is never wrong and the model never overrides it
  • Handles the hard cases: a bag of wet food → wet, a greasy container → rinse first, sharps → wrap and label
  • Searchable guide, animated best-practice tips, and a municipal accountability + rewards layer

React 19 · Vite · Express 5 · Neon Postgres · Prisma · Google Vertex AI (Gemini)

View live
Home-cooked food in steel tiffin boxes on an apartment kitchen counter

Padosi

Live MVP

Neighbourhood home-food ordering

Home cooks in apartment societies sell the day's food through noisy WhatsApp groups, with no menu and no clean way to order. Padosi gives each society one link: cooks post a single live, today-only batch; residents see what's still available, order in a couple of taps, and pay the cook directly by UPI, then get WhatsApp updates all the way through to 'ready'.

  • One society link opens today's live menu: order only what's left, no scrolling a group chat
  • A real order loop: place → pay (UPI / Razorpay) → WhatsApp confirmed → cook marks ready → status + review
  • Retention built in: win back lapsed regulars instead of shipping vanity analytics

Next.js 16 · Drizzle · Neon Postgres · Razorpay · Meta WhatsApp · Vercel

View live
Neatly stacked vendor documents, folders and a stamp on a desk

TrustLink

Live

Vendor management

Vendor onboarding is a manual slog of spreadsheets, email threads, and chased documents. TrustLink automates the whole workflow into one auditable flow: requirements auto-shortlist matching vendors, onboarding and document collection run themselves, compliance checks and approvals are routed automatically, and verified vendors flow straight into a shared directory with versioned contracts and renewal reminders.

  • Auto-shortlists vendors against a requirement, with no manual sourcing
  • Self-service onboarding and document collection with automated compliance checks
  • Routed approval workflow with auto-tracked contracts, versions, and renewals

React 19 · Express · Neon Postgres · Prisma · Better Auth · Tailwind

View live
A headset resting on a laptop in a quiet office

Convoy

In development

AI sales coaching

Coaching a sales team means a manager manually reviewing calls that mostly go unheard. Convoy automates it: AI reviews every call and turns it into concrete post-call feedback, telling reps what worked, what to fix, and how to handle the next one. The coaching loop runs on its own, so closure rates improve without a manager sitting on every call.

  • Automated post-call analysis on every call, so nothing goes unreviewed
  • Turns raw conversations into specific, actionable coaching automatically
  • A self-running feedback loop focused on conversation quality and closure

React 19 · Express · Neon Postgres · Prisma · Better Auth

Track record

₹1,500 Cr+

first-year sales impact from a product that automated the pricing process

20+ hrs

of manual analysis automated away every week at Freyr Energy

+80%

accuracy gain from automating 100 MB Excel crunching into insights

+35%

daily active users on a market-intelligence SaaS platform

₹50+ Cr

saved through plant-control automation at Indian Oil

  1. Freyr Energy

    Senior Product Manager

    Aug 2026 – Sep 2026

    • Built an AI skill that crunches 100 MB Excel files into insights, replacing a manual analysis process.
    • Saved 20+ man-hours every week and improved accuracy by 80%.
    • Owned 5+ PRDs end to end, from requirements gathering and solutioning through developer walkthroughs, UAT, and go-live.
  2. FutureBridge

    Solution Consultant

    Aug 2024 – Jul 2026

    • Owned the product roadmap for a subscription market-intelligence SaaS, prioritizing features from customer feedback, analytics, and competitive benchmarking.
    • Shipped AI-powered search, new dashboards, news spotlight, and personalization; grew DAU by 35%.
    • Carried a $330K individual revenue target within a $2.5M practice, and achieved 110% in FY25–26.
  3. KPMG India

    Solution Consultant

    May 2021 – Jun 2024

    • Designed and shipped a B2B/B2C pricing SaaS (web + mobile) for a ₹4-trillion company across six business verticals.
    • Ran primary interviews across business functions to map requirements and success metrics, then recommended interventions using a shareholder-value-maximization framework.
    • Built OCR-driven market capture, SAP integration, and an embedded approval workflow into the tool.
    • Shipped 5+ data-analytics use cases to refine costing methodology, with SQL-based analysis, ETL design, and SIT/UAT through to go-live sign-off.
    • Led a cross-functional squad of Java, Python, SAP (MM/SD), and database engineers across the build.
    • Delivered 12% sales-volume growth worth ₹1,500 Cr+ in the first year of implementation.
    • Post go-live, drove continuous enhancements across navigation, dashboards, market-intelligence platform integrations, and ML-based pricing recommendations, growing product usage 72% within a year.
  4. Indian Oil Corporation

    Production Engineer

    Aug 2016 – May 2019

    • Delivered ₹125+ Cr in production value through pipeline projects and saved ₹50+ Cr via control automation.
    • Cut permit cycle time 90% with an online work-permit system on SAP; 1000+ accident-free days.
    • Rated an “Outstanding” performer across the plant.

About

I pick a problem space, get close to it, and build with AI to find out what a real solution could look like.

I'm a product and solution consultant who has spent 8+ years across procurement, pricing, and market intelligence, sizing markets, shaping portfolios, and shipping SaaS with engineering and design teams. That work taught me how to spot a problem worth solving. Over the past few months I've been putting that to work on AI: I take a problem space and build a product end to end, from problem statement to production, to test what the technology can really do.

MBA
IIM Lucknow
2019–21
B.E.
IIT Bombay
2012–16
CFA Level 1
CFA Institute
2021
Project Management
Google
2023
Rohit Mahadev

Have a problem space worth exploring, hiring, or want to compare notes?

rohitmahadev10@gmail.com