Ishank Sharma

Forward Deployed Engineer

Los Angeles, California

The messy part
is my kind
of work.

I’m Ishank. I turn ambiguous problems into AI products that work in the real world. From the first customer conversation to the last mile of production.

Explore my work
A familiar starting point The messy middle of forward deployed engineering Scattered customer asks, people, data, legacy tools, risks, and ownership questions. Make it make sense arranges them into discovery, framing, building, validation, operations, and handoff. A vague askDiscover“Can we add AI?”Find the real need PeopleFrameDifferent prioritiesAgree what matters Legacy toolsBuildHidden dependenciesFit the environment Messy dataValidateWhat can we trust?Prove it in real use UnknownsOperateWhat if it breaks?Recover. Improve. Who owns it?TransferBeyond the demoA team can own it The real brief is rarely a neat brief. Learn in the field. Feed it back. Real peopleLegacy systemsA little AIA lot of unknowns

A little like my day job. Give it a click.

Customer context meets engineering depth.

Applied AIBackend systems0-to-1 ownership

Things I’ve
put into motion.

Useful software starts with a real problem.
Here are a few of mine.

Prentis AI / 2026–present

From possibility.
To product.

Forward Deployed Engineer, building AI-powered products from discovery to production.

  • Applied AI
  • Product engineering
  • 0-to-1
Three kinds of judgment. One working workflow.AI interprets, code checks, and people decide. Together they support a reliable workflow with evidence, feedback, and a team that can own it. A generic illustration, not a client system. AIInterpret CodeCheck PeopleDecide A workflow that works.Evidence. Recovery. A named owner.

A way of working. Not a client blueprint.

From1 day
To3 min

Same decision. A much shorter wait.

CommerceIQ / Data & infrastructure

Making the waiting disappear.

A media budget recommendation pipeline built with AWS and roughly 40 SQL modules. Processing fell from a day to about three minutes, with approximately $200K in annual client savings.

Step Functions, SQS, Lambda, SQL
Has anyone solved this before?

Let’s start with what
your team already knows.

Ella / Knowledge systems

A better memory for the team.

A Slack assistant that finds answers in company knowledge, escalates unanswered questions, and turns those answers into knowledge the next person can use.

View my GitHub

Also on my workbench.

Projects, experiments, and a few early builds that shaped how I work today.

6 projects in the collection

How I work

Close to the problem.
All the way
through.

The interesting part is connecting the pieces: what people need, what the model can do, and what the system must guarantee.

  1. 1

    Understand the actual work.

    Talk to the people doing it. Trace the workarounds. Find the problem beneath the feature request.

  2. 2

    Use AI where it earns its place.

    Models for ambiguity. Deterministic code for rules. Clear boundaries between the two.

  3. 3

    Own the last mile.

    Integration, evaluation, recovery, and adoption. The job is done when the workflow actually works.

A little about me

Engineer by trade.
Curious by default.

Ishank Sharma in San Francisco
Usually asking one more question.

I like problems that don’t arrive neatly packaged.

My foundation is in backend, data, and distributed systems. At CommerceIQ, I built pipelines, optimized infrastructure, and grew into Forward Deployed Engineering. At Prentis AI, I’m taking on the whole journey: discovery, product, engineering, and production.

I completed my M.S. in Computer Science at CSULB in May 2026. These days, I’m based in Los Angeles, working at the intersection of customer context and applied AI.

Outside the implementation details, I’m drawn to philosophy, psychology, and questions about how people think. There’s usually a good film somewhere in the mix, too.

Full profile & résumé LinkedIn
The journey so far

Built on experience.
Still asking questions.

From backend and data foundations to owning the whole path from customer problem to working product.

Open résumé
Jun 2026–presentPrentis AIForward Deployed Engineer

Building AI-powered products from discovery to production. Working across customer discovery, product engineering, and the realities of getting software into use.

Feb–May 2026CommerceIQForward Deployed Engineer

Built multi-tenant analytics experiences for executive teams, bringing together dashboards, causal analysis, and AI-generated insights.

Productionized Ally, an executive-summary system using GPT-4o with structured prompts and guardrails.

Sep 2025–May 2026California State University, Long BeachInstructional Student Assistant · Explainable & Ethical AI

Supported labs, grading, and student learning around model interpretability and responsible AI.

May–Jul 2025Stealth startupFull Stack Engineer

Built across a 0-to-1 product, including Text-to-SQL, semantic search, and LLM inference. Compared model quality, latency, and cost; added caching, fallbacks, and regression tests.

Feb–May 2025California State University, Long BeachResearch Assistant

Explored webcam-based gaze tracking and advertising effectiveness, including calibration, gaze heatmaps, visual attention, and saliency research.

Sep 2021–Aug 2024CommerceIQSoftware Engineer · SDE 1 → SDE 2

Built backend, data, and advertising systems. A media-budget pipeline reduced processing from roughly a day to three minutes and delivered approximately $200K in annual client savings.

Worked on cold-start recommendations, the migration of budget-prediction workloads from Snowflake to Databricks, Amazon Advertising systems, and Criteo strategy automation.

Feb–May 2021CommerceIQSoftware Engineering Intern

Investigated Snowflake multi-cluster concurrency and built monitoring dashboards to understand performance and tail latency.

Education

California State University,
Long Beach

M.S. Computer Science · 2024–2026
Completed May 2026 · GPA 3.9 / 4.0

Ramaiah Institute
of Technology

B.E. Information Science & Engineering
2017–2021 · Bengaluru, India

Some familiar tools.

Applied AI

Claude, Amazon Bedrock, GPT, RAG, embeddings, Text-to-SQL, evaluation, and human-in-the-loop workflows.

Backend & data

Python, Java, SQL, JavaScript, FastAPI, Flask, PostgreSQL, SQL Server, Snowflake, and Databricks.

Cloud & delivery

AWS, Docker, Step Functions, SQS, Lambda, CI/CD, observability, and systems that recover when things go wrong.

A competitive streak

A deadline.
A team.
A working idea.

Hackathons were an early way to practice the full loop: find a useful problem, build something real, and explain why it matters.

  • Smart India Hackathon

    Winner · Team Avocado

    Police Beats Allocation: tools for digitizing police operations, built for the National Crime Records Bureau.

  • Microsoft IncubateIND Innovation Series

    Runner-up

    Women Suraksha, using maps and emergency alerts to connect people with help.

  • CommerceIQ AI Hackathon

    Honorable mention

    Using language models to understand product-review sentiment and compare competitors.

  • GE Healthcare Precision Hackathon

    Honorable mention

    A project focused on hospital workflows and waiting times.

  • Reverie Language Hackathon

    Runner-up

    Exploring translation and multilingual communication.

  • IEEE Ideathon

    Runner-up

Thinking out loud · Selected writing

A few ideas
outside the code.

From data visualization to the fourth dimension. A few pieces from my earlier writing, and the questions I was working through.

More on Medium