Akshita Gandra

Senior product manager and co-inventor on a U.S. patent, building AI/ML, search, and platform products.

MS in Data Science from UC Berkeley, focused on NLP, RAG, and model evaluation. I ship my own products, and I write about responsible AI.

LinkedIn GitHub Resume

Hover or tap a card

Work

Five years across travel and ad tech.

U.S. Patent

12,621,508 B2

Co-inventor

Just-in-Time VOD Ad Architecture

Publica, acquired by IAS × Samsung

Video-on-demand ad breaks fail when the ad request lands too late to fill the slot. The architecture uses placeholder segments, pre-break fetch signals, and caching so the request resolves before the break begins. Modeled render rate improved from 39.6% to 81.8%.

View patent

Where the ad request fires

VOD streamAd break

Break

Placeholder segment

Ad request

39.6%Modeled render rate
request lands at the break

Pre-break fetch signalPlaceholder segmentsCaching

Selected impact

Spotnana Senior Product Manager · 2024 – present

Product vision across Homepage, Checkout, Trips, and Hotel & Ground. Sole PM for Hotel & Ground, covering search, ranking, checkout, and supplier integrations, plus conversational AI and multi-agent travel workflows. Owned the Marriott client relationship and launched the first direct connection of its kind in the industry.

$374MHotel spend, up 80%

+75%Bookings growth, year over year

15+Engineers and designers aligned across global teams

Supplier integrations launched

Marriott Booking.com Travelodge Premier Inn GroundSpanand more

Also shipped

Self-service rail changes Workflows and Travel AI

Publica, acquired by IAS Product Manager · 2021 – 2024

CTV and programmatic product across ad measurement, fraud detection, server-side ad insertion, and platform APIs. Shipped the Bidders, Orders, and Trafficker APIs and co-led Platform Connect.

$1.5M+Audience Manager, taken 0 to 1 and scaled to ARR

+90%SSAI revenue, $833K to $1.58M

$20M+Annual revenue supported by products owned

Intel Program Manager Intern · 2020

Built Intel's first internal ML enablement platform and drove adoption through self-service training and documentation, with a VP-level ML readiness report.

10ML initiatives supported

13Cross-org projects onboarded

Research

Berkeley and Davis.

MS, Data ScienceUC Berkeley · 2024 – 2026NLP, RAG, model evaluation, and large-scale data systems.

BS, Computational Cognitive ScienceUC Davis · 2017 – 2021AI/ML, computer science and programming, linguistics, psychology.

Applied NLP

Detecting Suicidal Ideation in Text

BERT · RoBERTa · BERTweet / 232K Reddit posts

Fine-tuned and compared three transformer architectures against a baseline on 232K labeled posts. The interesting part was not the headline score. I used SHAP and failure analysis to study false negatives, the errors that matter most in this setting, and outlined what clinical deployment would actually require.

Read the paper Code

0.994

RoBERTa, F1 on the risk class. Best of three architectures compared.

Focus areasNLPRAGModel evaluationRanking & personalizationAgentic orchestration

Building

Started as my Berkeley capstone. I rebuilt it solo and shipped it.

Shipped product · Full-stack AI

Clear Cart

Livetheclearcart.com

Food recall and ingredient scanner. Scan a receipt, barcode, or photo of your cart to check FDA recalls and flag ingredients against your preferences. Next.js, Supabase, GPT-4o vision.

Began as Safe-Cart, my UC Berkeley MIDS capstone with a team of five. I rebuilt it from scratch on my own afterward and shipped it as a real product.

Try it out Berkeley project

Capstone code

clearcart · scan result

SCANGranola Clusters, 12 ozUPC 0 41220 10056 8 · matched

RecallActive FDA enforcement report on this lot

Tree nutsFLAGGED · your profile

Soy lecithinOK

Cane sugarOK

Representation of the product surface