A product mind with builder's hands.

I do the product thinking and ship the code. I find a problem worth solving, get to the bottom of it, and turn it into a working product you can click, often in a matter of days.

The product side

Figure out what to build

Customer interviews, positioning, prioritization, the go/no-go case. The judgment about what's worth building, and why.

The builder side

Then actually build it

Full-stack and hosted, with real integrations and AI under the hood. Not a slide deck, the thing itself.

01  The short version

Four years doing product. Five years building.

Not new to either. I started as an engineer, moved into product, and never stopped writing code, which is why I can do both halves of the job today.

Product~4 years
2022 to now
Building~5 years
2021 to now
202120222023202420252026

Engineering came first (Flask apps, automation, a shipped data tool). Then product: analyst, then product manager, across a startup, an Nvidia-backed company, and a Fortune 500. The two have run together ever since, and lately I've fused them, shipping full products solo.

02  The builds

Six problems. Six working answers.

Mostly built solo in a single continuous sprint, end to end, and live right now. The time it took is right up front. Open any to see the technical workflow.

~12 hrsidea to live, one sprint
Search visibility for the AI era
AEO Audit Tooldoes an AI name you?
LIVE

It tells a company whether ChatGPT and Gemini name them when buyers ask, and scores them red, yellow, or green. Search is moving from Google to AI engines, and whoever the AI names in its answer wins the customer. No one had a way to measure it, so I built one.

The hard callRule-based scoring, not asking an AI to grade. Asking GPT-4 each time is slower, pricier, and answers differently every run. Rules make it free, instant, and identical every time.

see the technical workflow
How a scan runs, end to end
01Take a real buying question like "best office space platform in SF" and fan it out to both AI engines at once.
02Call GPT-4o and Gemini in parallel, each with live web access, so total time is the slower of the two (~8s), not the sum (~16s).
03Score each answer with rules, matching company names, addresses, and pricing against curated lists and patterns. Zero AI cost, millisecond result, identical every run.
04Parse the page's hidden structured dataThe machine-readable labels in a page's code that tell an AI what a business is and sells. Most sites get this wrong. from raw HTML, tolerating broken markup instead of crashing, to see what an AI engine can actually read.
05Return a red / yellow / green scoreboard plus the exact <head> code snippet to fix what's missing.
Next.jsTypeScriptGPT-4oGeminitype-safe API
~38 hrsidea to live, one sprint
Hiring quality
AI Recruiter Coachinterview quality, made visible
LIVE

It turns recorded recruiter interviews into automatic coaching and a quality score managers can finally see. The data sits in every call, and nobody was turning it into feedback, so I built the system that does.

The hard callFully automatic, but with a human in the loop. Everything runs itself the second a call ends, yet a person reviews before anything reaches the recruiter, because hiring is too sensitive to fully automate.

see the technical workflow
From finished call to coaching
01A call ends and fires a webhook, so the system reacts instantly with no manual upload.
02Pull the recording and transcribe it, storing a timestamped transcript.
03Run an AI evaluation for an interview score, strengths, and improvement areas, and flag any sensitive or non-compliant questions.
04Route to the hiring lead to review before the recruiter sees it, then track how each recruiter improves over time.
Next.jsPostgresOpenAIwebhook capture
~7 hrsidea to live, one sprint
Full-stack proof
Image Transform Serviceempty repo to hosted product
LIVE

Upload an image, get back a processed, hosted version with a link to share. A clean test of taking something from empty repo to a tool that feels finished, not a coding exercise.

The pointEnd to end, with the polish that says finished. Every loading and failure case handled, live and hosted, code public. Here's the link, go press the button.

see the technical workflow
Upload to shareable link
01Accept an image (PNG, JPEG, or WEBP) with clear progress feedback.
02Run the pipeline: automatic background removal, then a transform.
03Host the result and return a shareable URL, with the option to regenerate, swap, or delete in-app.
full-stackimage APIhosted storageshareable links
4 monthsongoing strategy work
Positioning & product critique
NextLMsharpening an AI product
LIVE

A product teardown of an early-stage AI sales tool: how to position it, and what's broken under the hood. I went at it as a user trying to break it and a product person trying to sharpen it.

The insightIt's a timing advantage, not just another lead tool. Its real edge was telling you when a buyer is ready, but the homepage led with generic "find leads" language, so skimmers filed it as ordinary. I made the case to lead with the timing edge.

the product eye, real things a busy team missed
  • Gated data leaking out through an export path.
  • A login flow reporting success for email addresses that were never registered.
  • Data that claimed to be current but didn't match reality, all found by using the product carefully and writing it up clearly.
positioningonboardingsecurity findings
~5 daysincl. becoming a partner
Go-to-market tool
AgentForce AlignerI became a partner to build it
LIVE

It connects to a company's Salesforce, scores how ready they are for AI agents, and projects the return on adopting. Salesforce's adoption push wasn't landing; two interviews told me why, and the fix was leading with numbers.

The moveI registered as an authorized Salesforce ISV partner org just to build it. Not a mockup: a real connect-your-account-and-see-your-score tool, built solo from idea to working prototype in five days.

see the technical workflow
Connect to ROI projection
01A rep connects the company's Salesforce with a secure account link in seconds.
02Read the org's data and score readiness, surfacing exactly what's too messy for the AI to perform on.
03List the specific fixes that would make the AI agents shine, then project the ROI of adopting so the decision leads with numbers.
Pythonsecure account connectOpenAIofficial partner build
~26 hrsidea to live, one sprint
Sales automation, end to end
Competitor-to-Outreach Enginefind their clients, draft the email, all but send
LIVE

A founder wanted his competitors' clients so he could pitch them, and was finding them one by one through LinkedIn mutual connections. Painful, slow, and incomplete. I replaced the whole thing with one pipeline: it finds the companies, finds the right decision-maker at each, and writes a tailored cold email, so all he does is press send from his own inbox.

The leapFrom a manual LinkedIn crawl to a one-button workflow. The hard part wasn't any single step, it was chaining scraping, AI structuring, contact enrichment, and email drafting into one flow that ends with a ready-to-send draft sitting in his inbox.

see the technical workflow
Competitor name to ready-to-send email
01Crawl a competitor's site and public footprint and pull the raw content that hints at who their clients are.
02Use the OpenAI API to structure the mess into clean target-company records, the step that turns scattered HTML into a real list.
03Find the right decision-maker with Apollo, enriching each company with the exact person, their title, and a verified work email.
04Generate a tailored cold email per contact with the OpenAI API, written to the specific person and company, not a template blast.
05Drop each draft straight into his inbox, addressed and ready. The only human step left is pressing send.
PythonOpenAI APIApollo.ioweb scrapinginbox integration
03  The professional track

And I've done product work inside real companies.

A Fortune 500 manufacturer, an AI startup backed by Nvidia, and a software studio building for clients worldwide.

Textron
CURRENT ROLE
Product Manager Intern, Product Strategy
Jun 2026, present
United States

Product strategy at a Fortune 500 manufacturer, on a new vehicle program where hardware and software meet. (Some specifics are confidential, so this stays high level.)

  • Leading the business caseA structured argument for whether the company should invest in building a product: market, competition, what customers want, and whether the numbers justify it. for a new product concept: sizing the market, mapping the competition, capturing what customers want, and shaping the value proposition behind a major capital-investment decisionWhether the company commits significant money to develop the product. The kind of call that needs evidence, not a hunch.. Genuine zero-to-one work.
  • Acting as the product-side bridge between in-house engineering and an external software partner, turning business needs into clear requirements and closing the gaps between intent and what gets built.
  • Partnering on a data product that turns field sensor data into early-warning signals, helping prevent costly downtime across a fleet.
EmeraldAI
BACKED BY NVIDIA
Student Product Consultant
Aug 2025, May 2026
Washington, DC

Market and competitive strategy for a $34.5M AI startup working where data centers meet energy.

  • Ran market and competitive research across three global regions to find expansion opportunities, fluent in a brand-new market within a week.
  • Built the model weighing battery capacity against AI-driven energy savings, feeding the international expansion plan.
  • Set early product direction across both software and hardware (energy management and battery integration for data centers).
Crayond Digital
Product Analyst, then Associate Product Manager
Apr 2023, Apr 2025
Chennai, India

Full product ownership across a multi-year run, shipping software for clients worldwide.

  • Started as Product Analyst: owned research on a product across 57 competitors and cut scope 30% through user research that focused the roadmap.
  • Promoted to APM: co-led a 23-person team building eight point-of-sale apps for the Middle East market, wrote 450+ product specsPRDs, BRDs, and MRDs: the documents that define what gets built, why, and for whom. The blueprints engineering works from., coordinated 24 stakeholders, hit 96% on-time delivery and cut defects 20%.
Tactii
Product Analyst Intern (first PM hire), earlier Developer Intern
2021, 2022
Remote

The bridge between my engineering background and product: I started building, then moved to defining what to build.

  • First PM hire: set up the team's planning from scratch and improved planning efficiency ~85%, taking two products from zero to one.
  • Earlier, as a developer, built and shipped a data-visualization tool (Flask) used by real users, plus scraping and automation pipelines.
04  How I work

Same approach, every time.

I don't write memos about problems. I build the solution and bring it to the table.

First

Find the real gap

A problem that's genuinely unsolved. The signal is usually already there; most teams just can't see it yet. I find it by talking to real people.

Then

Get deep, fast

Understand it cold. Interviews, public data, fluent in a new field within days. Then choose the one technical decision that makes it work.

Finally

Ship the answer

Full-stack, hosted, something you can click. Built fast with AI in the loop, so deciding to shipping is days, not months.

05  Get in touch

Give me a real problem. I'll bring back a working answer.

A product mind and builder's hands in one person, from figuring out what to build to shipping it. I'd rather show you the thing than describe it.