Product Manager AI & Document Intelligence San Francisco

I teach software to read the documents nobody wants to.

I build AI agents that turn enterprise content nobody has time to read into answers people actually act on. Before product, I wrote Swift for a living — which is why I still open the network tab when a PM probably shouldn’t.

An illustration of a document-extraction interface reading Aditya's profile and pulling out structured fields.
sreerama_aditya.profile extracting

Aditya Sreerama is a Product Manager at Egnyte in San Francisco, where he led the 0-to-1 launch of a Contract Analyst AI agent now used by 50+ enterprise customers to analyse 40,000+ contracts. He ships retrieval, evaluation and extraction systems — and started out writing Swift for an airline.

role Product Manager
company Egnyte
location San Francisco
customers 50+
contracts_analysed 40,000+
origin_story Swift

Yes, I built the thing that does this.

161,000+ users on the homepage I rebuilt
40,000+ contracts run through my agent
27,000 customer domains reached
95+ user interviews, and counting
$50M+ in assets unlocked at Marstone

Selected work

Four things I shipped that were harder than they looked

Mostly at Egnyte, where the job is convincing a decades-old content platform to behave like an AI-native one without breaking a single customer's Tuesday.

0 → 1 · AI Agent

Contract Analyst

An agent that reads a contract the way a paralegal would — pulling renewal dates, liability caps and indemnity clauses out of 21 contract types without anyone opening the PDF. I owned model selection, prompt architecture and the evaluation framework that kept extraction precision honest against cost.

40,000+
contracts analysed
50+
enterprise customers
75+
discovery interviews
  • LLM evaluation
  • Prompt architecture
  • Extraction
  • 0→1

Retrieval · Multi-modal

Search that answers instead of listing

Replaced keyword search with RAG-backed summaries and multi-modal retrieval across video, audio and geo content. The measurable win was behavioural: people stopped guessing at filenames and started asking questions in plain English.

RAG
summaries at query time
3
new modalities indexed
  • RAG
  • Search UX
  • Multi-modal
  • Behaviour change

Vertical · Document AI

Construction documents, decoded

A pipeline that classifies RFIs, RFPs, submittals and drawings the moment they land, then extracts the fields that only matter in construction — CSI divisions, project sections — so downstream AI answers arrive with the right context already attached.

4
document types classified at upload
CSI
divisions extracted automatically
  • Classification
  • Vertical AI
  • AEC
  • Metadata

Platform · Adoption

The homepage nobody had loved in years

The first screen every user sees had quietly become a filing cabinet. I rebuilt it as a personalised starting point — recent work, relevant content, actual signal — and shipped it carefully enough that it became the default for everyone.

161,000+
users adopted
27,000
domains
Default
for all new customers
  • Personalisation
  • Migration strategy
  • Platform

How I work

Five stages, and the discipline to not skip the first one

Most bad products are just good execution pointed at the wrong problem. So I spend longer than most people find comfortable at the start.

  1. 01

    Discover

    Find out whether the thing should exist before anyone commits a sprint to it.

    • Opportunity briefs
    • Competitive analysis
    • Personas & segmentation
    • User research
    • Journey mapping
    • Customer interviews
  2. 02

    Design

    Make it legible on paper first. Cheapest place to be wrong.

    • Wireframing
    • Information architecture
    • Data-driven design
    • Figma Make
  3. 03

    Prototype

    Arguing about a mockup takes longer than clicking a real one. So I build the real one.

    • Claude Code
    • Cursor
    • Antigravity
    • Balsamiq
    • POC development
  4. 04

    Build

    Where the roadmap meets engineering reality, and reality usually wins.

    • Product roadmap
    • Iterative development
    • User stories
    • Jira
    • Agile
  5. 05

    Launch

    Shipping is the start of the work. Nobody adopts what nobody understands.

    • Go-to-market strategy
    • Sales & support enablement
    • Marketing material

The long version

Engineer in, product manager out

Six years of shipping, in reverse. Start at now and scroll back to the version of me who thought a spec was a suggestion.

  1. Now

  2. Nov 2023 — Now San Francisco Bay Area

    Product Manager at Egnyte

    AI-powered content platform for secure file management, collaboration and governance.

    Hired to work on content management. Ended up spending three years teaching a filing system to think.

    • 40,000+ contracts analysed by the Contract Analyst AI agent I took from zero to launch, across 21 contract types and 50+ customers.
    • +25% answer accuracy from an SLM-powered metadata enrichment strategy that infers meaning from folder hierarchy instead of asking users to tag anything.
    • 161,000+ users on a rebuilt, personalised homepage — now the default experience across 27,000 domains.
    • Shipped RAG-based summaries and multi-modal search across video, audio and geo content, moving people from keyword search to asking real questions.
    • Built an AEC document intelligence pipeline that classifies RFIs, RFPs, submittals and drawings at upload, then extracts CSI divisions and project sections downstream.
    • LLM evaluation
    • RAG
    • Prompt architecture
    • SLMs
    • Enterprise content
  3. Feb 2023 — Jun 2023 New York, NY

    Product Manager at Marstone

    Lets financial institutions launch white-labelled wealth management.

    Digital wealth management, where the users were credit unions and the stakes were other people’s retirement.

    • $50M+ in assets under management unlocked by onboarding two credit unions onto the Digital Advice Platform.
    • +20% session duration after redirecting UI/UX requirements for the account activity and maintenance pages.
    • 20+ user interviews that pointed goal-setting features squarely at what Gen Z actually wanted.
    • Fintech
    • Roadmapping
    • User research
    • UX direction
  4. Jun 2022 — Dec 2022 Chatham, NJ

    Product Manager, Growth Analytics at DriveWealth

    API-powered brokerage letting fintechs embed US securities trading.

    Walked into a company with no product analytics and left with dashboards three teams refused to give up.

    • 10x faster reporting after building product analytics and visualisation from scratch for Product, Sales and Engineering.
    • $50M+ in product revenue tracked, via growth dashboards and a repeatable process built with a cross-functional team of four.
    • $3.8T market analysed — Gen Z fintech research in Tableau, turned into B2B2C feature concepts and Figma wireframes.
    • Tableau
    • Growth analytics
    • Figma
    • Market research
  5. 2021 — 2022 Durham, NC

    M.Eng. Engineering Management at Duke University

    Product Management & Entrepreneurship track.

    The deliberate pivot: engineer in, product manager out. Also where I learned that a good framework beats a loud opinion.

    • Master of Engineering Management — product management, marketing strategy, competitive strategy and digital transformation.
    • Led the team behind a tuberculosis care app that reached the global top 3 at Microsoft’s Imagine Cup 2023 and won the Epic Challenge.
    • Product strategy
    • Marketing strategy
    • Digital transformation
  6. May 2020 — Aug 2020 Bangalore, India while at AirAsia

    Founder & Product Lead at RemotePrint.tools

    Contact-less print and delivery for university students during the pandemic.

    Lockdown shut every campus print shop, so a thousand students had nowhere to print their exams. We fixed that in a week.

    • 1,000+ students served by a web platform a team of four designed and shipped in seven days.
    • 50,000+ sheets printed through two vendor partnerships I negotiated on service and quality terms.
    • $3,500+ in revenue generated in two months — profitable, and then happily obsolete.
    • 0→1
    • Ops
    • Vendor negotiation
    • Marketplace
  7. Jan 2020 — May 2021 Bangalore, India

    Software Engineer, iOS at AirAsia Technology Centre

    Southeast Asia’s largest budget airline.

    Where I learned what it costs engineering to build the thing a product person casually asked for on a Tuesday.

    • +15% daily app opens within two months of integrating the CleverTap module for targeted push and offers.
    • +70% login persistence from a unified Swift authentication module bridging native and web identity.
    • 4–8x faster development cycles after automating QA and localisation with Python and Google Apps Script.
    • Swift
    • Python
    • Google Apps Script
    • iOS
  8. 2016 — 2020 Bangalore, India

    B.E. Information Science & Engineering at RV College of Engineering

    Information Science & Engineering.

    Four years of computer science, and the beginning of a long habit of building things nobody asked for.

    • Computer science
    • Information systems
  9. Bangalore, a laptop, and too much free time

Say hello

Got a hard product problem?

I’m most useful early — when the problem is still fuzzy, the data is a mess, and someone needs to decide what’s actually worth building. Tell me about yours.

Usually replies within a day. Always replies to a good problem.