AI Engineer

Subhan Shaikh — AI Engineer. Building the layer between LLMs and real-world use.

Anyone can prompt an LLM. I build the layer that makes it reliable, measurable, and worth deploying.

View my work Download résumé
subhan@portfolio — zsh
➜ ~

About

Subhan Shaikh

Mumbai to California is a long way to go to debug a retrieval pipeline. But that’s roughly how it went.

IBM gave me my first real AI project. UCSC gave me a 4.0 and over a year of research in explainability and Vision Language models, the kind of research that changes how you think about building, not just what you build. Gogentic gave me a startup environment, a small AI team, an eight-week deadline, and an infrastructure cost problem I cut by 80%. CalHacks 2025 gave me three sponsor track wins.

What I got out of all of it is harder to put on a resume. I know where these systems break, and I know how to make sure they don’t.

Experience

  1. Applied AI Engineer · Gogentic.ai

    Remote · Jun 2025 – Nov 2025
    • ─Led the AI team to deliver a fully functional AI-driven backend MVP within an 8-week sprint, supporting on-schedule user testing with zero major production incidents.
    • ─Reduced AI infrastructure costs by 80% by replacing third-party APIs with evaluated open-source models selected against task-specific quality benchmarks.
    • ─Boosted backend service performance by 50% through custom inference workflows and strategic pipeline optimization.
    80%
    cost reduction
    50%
    perf boost
    8wk
    to MVP

  2. AI Automation Intern · Propel Flow

    San Francisco Bay Area · Jun 2025 – Aug 2025
    • ─Built AI-driven automation tools for community interaction analysis and response across LinkedIn, Slack, Facebook, and Reddit.
    • ─Analyzed community interactions to surface customer pain points and shipped automated solutions to resolve them at scale.
    • ─Collaborated cross-functionally to ship features that improved user engagement and service efficiency.

  3. Graduate Researcher · UC Santa Cruz — AIEA Lab

    Santa Cruz, CA · Jan 2025 – Mar 2026
    • ─Researched advanced Retrieval-Augmented Generation techniques to build a CS-department specific chatbot over course and program data, under Professor Leilani Gilpin's AI Explainability & Accountability Lab. Explored Prolog-based symbolic reasoning as part of broader research into LLM explainability.

  4. Course Tutor, CSE 40 (Intro to ML) · UC Santa Cruz — Baskin Engineering

    Santa Cruz, CA · Sep 2025 – Mar 2026
    • ─Tutored students in foundational machine learning, reinforcing core concepts and problem-solving over exam prep.

  5. AI Intern · IBM

    Remote · Sep 2022 – Nov 2022
    • ─Built a sign-language recognition system end-to-end on the ASL dataset, achieving 90% classification accuracy across the full pipeline from preprocessing to deployment.
    • ─Delivered a working prototype and documented findings for a stakeholder presentation.
    90%
    accuracy

Featured Work

★ currently building

Articulr

A personal branding assistant on autopilot. Turns a professional's real experience into consistent personal-brand content that compounds over time — no blank page, no writing skills required, published in their own voice.

LLMsAgentsNext.js
★ CalHacks 2025 Winner

Orbit.ai

A modular, MCP-agnostic developer productivity agent with real-time workflow automation and natural-language voice command execution. Won three sponsor tracks at UC Berkeley's CalHacks 2025, the world's largest collegiate hackathon.

PythonMCPMulti-agent

RepairMate — AI-Driven Home Device Diagnostics

Proprietary Lab Project

A multi-modal, multi-agent system that diagnoses household device faults from user-submitted video, fusing frame extraction, transcription, and Vision-Language model analysis into structured technician summaries. Automated prompt engineering with GEPA (a reflection-based optimizer) tuned against BERTScore. Applied model quantization for real-time on-device performance.

· 87% smaller model· 70% faster inference· 85% fault detection
AutogenOpenCVWhisperVLMs

Critique-Enhanced RAG — Self-Refining Retrieval-Augmented Generation

A self-critiquing RAG system that scores its own answers across four quality dimensions and conditionally refines those below threshold. Benchmarked three architectures (vanilla, LLM reranker, critique-refinement) across quality, cost, and latency.

· 27% quality lift (150 SQuAD)· refined 56% of answers
PythonOpenAI API

Rolofy — Digital Identity Platform (Live)

A full-stack networking platform that replaces manual handle-swapping with a single scannable identity, letting people instantly exchange all their social and professional profiles at events. Solo-built and deployed end to end.

· 2k+ impressions· 40+ user profiles
Next.jsFastAPISupabaseVercel
→ rolofy.online

Other Projects

Memora

View Work

An AI mental-health support chatbot combining conversational memory, voice interaction, and sentiment analysis.

Content-Crew

View Work

A multi-agent content generation system that collaboratively creates blog posts, images, and social content.

AutoCents

View Work

A car price prediction model that estimates a vehicle's current value from input parameters.

Achievements

★ CalHacks ’25

CalHacks ’25 — 3 Sponsor Tracks Won

UC Berkeley CalHacks 2025, the world’s largest collegiate hackathon

Built Orbit.ai, a modular MCP-agnostic developer productivity agent with real-time workflow automation and natural-language voice command execution. Won 3 sponsor tracks.

★ Ideathon Circuit

1st Place × 2 — Intercollegiate Ideathon Circuit

Mumbai, India · Thakur College of Engineering & Technology & K.J. Somaiya College

Won first place at two separate intercollegiate ideathons with a single idea — an AI-powered maternal wellness platform for expectant mothers in nuclear families, combining health assessments, personalized nutrition planning, ovulation tracking, and a culturally sensitive community forum. Two competitions, two first-place finishes.

Skills

Languages
PythonJavaScriptTypeScriptJavaCSQL
AI / ML
RAGMulti-agent systems & orchestrationMCP serversLLM fine-tuningModel evaluationInference optimizationModel quantizationPrompt engineeringVector databases
Frameworks & Libraries
PyTorchLangChainLangGraphLangSmithAutogenCrewAIHugging FaceFastAPINext.jsReact
Infrastructure
PostgreSQLSupabaseDockerGCPVercelGitCI/CDREST APIs

Education

MS, Computer Science — Master’s

Graduated 2026
University of California, Santa Cruz · GPA 4.0/4.0

Relevant coursework: Intro to AI, Machine Learning, Deep Learning, Projects in AI, Database Systems, Data Structures & Algorithms, Statistical Data Analysis.

BE, Computer Science — Bachelor’s

University of Mumbai, India

Contact

Currently open to SWE/ AI/ FDE engineering roles at ambitious teams.

If this resonates, LinkedIn is where to find me. GitHub is where to find the work.

subhan.shaikh.me@gmail.com
GitHub LinkedIn