CS @ UC IrvineGPA 3.92
Open to SWE/ML internships — Summer 2027. Applying to MS programs for Fall 2027

From product vision to computer vision

Full-stack engineer and product designer. I own products end to end, from drafting spec docs and testing prototypes to iOS, Android, and Web releases. Algorithmic design and problem solving is a strong suit of mine, fueling the efficient backends behind my builds. My technical foundation draws on four years in AI, senior CS coursework, projects for department faculty, and two computer vision publications.

now.ts
Rohan Singh
Rohan SinghUC Irvine ’27
Software · Machine Learning
0+
Shipped Products
0
Publications
0.00
GPA at UC Irvine
0+
Years in ML/CV

Experience

Sept 2026 – Present · Irvine, CA

Undergraduate Student Researcher

UC Irvine · Donald Bren School of ICS
  • Research in computer vision and computational geometry under Prof. Gopi Meenakshisundaram, working alongside the lab’s PhD students.
  • Building research tools for the lab: pipelines that run vision models and geometry processing over its data.
June 2026 – Present · Cupertino, CA

Full-Stack Engineer

Imentiv AI
  • Led PitchSense from brief to release: Imentiv’s sales-call intelligence platform, with three products on one FastAPI + Next.js pipeline — Pitch Coach sharpens a rep’s delivery, Reaction reads the audience, and Full Call shows how one moves the other.
  • Designed the scoring behind them: five engagement and six delivery constructs, each normalized to the speaker’s own resting baseline so a reserved rep isn’t graded on a showman’s scale, plus transcript-derived mechanics like fillers, hedges, pace, and pauses. The AI coach explains only those numbers.
  • Built A/B testing end to end: reps compare pitch variants on real viewer reactions across about 40 signals, with bootstrap 95% confidence intervals and minimum-sample guardrails before any winner is called, at no extra analysis cost.
  • Cut about 120 per-frame API calls from each analysis by reusing frames already fetched, and built a harness that meters every Imentiv call, its latency, and its credit cost, which showed where else to save, like skipping speaker matching on solo clips.
  • Designed SceneSense, a 0→1 acting coach that grades a take against a sentence-level map of the script’s emotional arc and tells an improvised line from a forgotten one; now building it.
March 2026 – Present · Irvine, CA

Founding Designer & Full-Stack Engineer

Merge Campus
  • Product design: Originated Merge as a carpool app and widened it into a social platform for real-life plans with friends, then defined it through student persona-based testing, ranked use cases, feature specs, and a 62-screen clickable prototype.
  • Architecture: One of two engineers (450+ of the team’s 1,150+ merged PRs) and the lead on system design for a Flutter + Firebase app with 110+ TypeScript Cloud Functions, from half its data model to the access model where the server, not the app, decides what each person sees.
  • Pipelines: Built a Gemini place recommender that plans its own Google Places searches, then grounds and verifies every pick against the results; receipt scanning into itemized bill splits; a travel-time and timeline solver; route optimization for carpools; and detour-based matching for peer-to-peer rides.
  • Security & quality: Wrote most of the Firestore security rules and about three quarters of the app’s 1,600+ test files and set up its first CI; fixes land test-first, re-verified by deliberately breaking them.
June 2025 – Sept 2025 · Cupertino, CA

AI/Machine Learning Intern

Imentiv AI
  • Boosted face detection and recognition accuracy by 6% via an automated ground-truth validation pipeline, directly optimizing production YOLO models.
  • Implemented a face re-identification mechanism using dynamic placeholders to maintain consistent tracking, validated against ByteTrack and BoT-FaceSORT.
  • Conducted comparative evaluations that influenced the engineering team’s decision to migrate from legacy models to updated YOLO architectures.
June 2024 – Sept 2024 · Cupertino, CA

AI/Machine Learning Intern

Imentiv AI
  • Engineered an image deduplication pipeline, using DINOv2 and Faiss to filter 95% similarity redundancies.
  • Developed an algorithm to identify and extract distinct video frames for dataset diversity.
  • Built a facial analysis tool using YOLOv8 to detect faces, create bounding boxes, and label emotions.
June 2023 – Aug 2023 · Remote

Software Engineer Intern

Asterbyte Software Systems
  • Designed and implemented a machine learning solution for emotion detection on human faces in videos and images.
  • Generated synthetic training data using Unreal Engine MetaHumans to improve model robustness.
  • Conducted research on emotion recognition and published findings at WCSET 2023 and IRAJ 2023.
June 2022 – Aug 2022 · Remote

Android Developer Intern

Asterbyte Software Systems
  • Engineered mobile application “Hue Christmas” for Philips Hue and reconstructed/optimized “Pitch Prime”.
  • Translated Pitch Prime from Spanish to English, expanding accessibility to a wider audience.

Projects

Full-stack · Multimodal AI

Behavioral Interview Coach

Soft-skills-first interview simulator that quantifies subconscious cues like facial expression and vocal tone to improve delivery.

  • Architected a ‘Soft-Skills First’ analysis engine designed to quantify non-technical traits (confidence, clarity, resilience, engagement), filling the critical feedback gap left by technical platforms like LeetCode.
  • Created a real-time interview simulation replicating high-pressure screenings in a low-stakes environment.
  • Developed a multimodal pipeline that separates semantic content from delivery when evaluating performance.
  • Gemini streams résumé-tailored questions for any company and role, then turns your session history into a personalized coaching plan.
PythonFastAPIImentiv APINext.jsDockerAWS S3GeminiComputer Vision
Full-stack · Generative AI

SikhAI

Full-stack spiritual companion: a scripture-grounded AI chatbot, a Punjabi ↔ English translator, Shabad search, the daily Hukamnama, and a community seva board.

  • Developed a Gemini-powered chatbot to provide spiritual guidance rooted in Sikh teachings.
  • Engineered a full-stack community platform featuring Firebase Authentication for a Seva volunteer event organizer.
  • Built an accessible, high-performance Shabad search for the Sikh community using Next.js.
  • Production-grade GenAI: schema-constrained JSON for the translator, graceful Cloud Translation fallback, and nonce-fenced prompts that block injection.
Next.jsTypeScriptGeminiFirebaseTailwind

Now building

Four products in the works. Merge Campus is a social platform launching at UC Irvine — and three sit on one thesis, delivery is a signal you can measure, each pointed at a different room: the stage, the sales call, the hard conversation.

iOS & Android · Social + AI

Merge Campus

Where plans with friends come together. Merge Campus is a social platform built around doing things in real life — exploring your college campus, managing plan logistics and feasibility, both timewise and carpool wise, coordinating a bring list between group members and the correlated expenses, along with many other quality of life features meant to support users in making their ambitious outings come to life.

  • The Timeline: A shared calendar finds timings that work for everyone involved. Planners can map out their outing stop by stop, while other members can see the plan first and join if it suits them. Makes it possible to view how adjusting the plan impacts the rest.
  • Expenses: Merge turns a scanned receipt into an itemized split, and offers running tabs between friends. There are options to divide a split by item, by serving, equally, by percent, etc.
  • Move Suggestions: Merge answers the classic “What’s the Move?” by suggesting locations in the area catered to the group’s interests, among several other adjustable parameters. It also suggests activities that can be done at home for fun, outlining any materials/items required to do an activity.
  • Getting there: Merge groups carpools, optimizes routes, and solves the carpooling concerns to minimize time spent driving unnecessarily.
  • Peer-to-peer ridesharing: Merge generates the detour distance and converts it into the added fuel cost, matches people based on route similarities, and provides them a platform to negotiate and decide what will work for their carpool purposes.
FlutterDartFirebaseCloud FunctionsGeminiGoogle Maps
iOS & Android · coming soon
0→1 Product · Multimodal AI

SceneSense

Imentiv AI · 2026

An AI scene partner for actors: it checks a take against what the script actually calls for — face, voice, and words — and knows an improvised line from a forgotten one.

  • Greenfield at Imentiv AI, from problem framing to scoring model: on-book mode coaches line by line, off-book returns a full post-take report.
  • The script becomes the rubric: a sentence-level map of up to 32 emotions, valence, and arousal, with every miss timestamped.
  • Separates improv from a dropped line: words and emotion still have to match, while a tempo drop or flash of confusion exposes a blank.
PythonImentiv APISpeech-to-TextEmbeddings
Full-stack · Emotion AI

Pitch Coach

Imentiv AI · 2026

Delivery coach for sales reps, designed measurement-first: the pipeline computes the evidence, and the AI is only allowed to explain it.

  • Grounded by design: every risk cites a number counted from the transcript, and the coach answers only from the computed timeline.
  • Six delivery constructs are normalized against the speaker's own resting baseline, so a reserved rep is never graded on a showman's scale.
  • One FastAPI pipeline serves three products: a scope flag turns the same analysis into Pitch Coach, Reaction, or Full Call.
PythonFastAPINext.jsPostgreSQLAWS S3GeminiImentiv API
Real-time · Multimodal AI

Clarity

LA Hacks 2026 · team build · architecture & pipeline design

A live AI conversation partner that listens the way people do — face, voice, and words together — because a text-only LLM misses most of what human communication actually carries.

  • LangGraph runs the turn loop, Gemma plays the other side of the table, and ElevenLabs streams that voice over WebSocket as it renders.
  • Catches what a text-only model cannot: each turn is read across face, voice, and language, so feedback covers tone and expression, not just wording.
  • Designed, built, and demoed in one weekend at LA Hacks 2026, then hardened ever since with correctness fixes and an offline test suite.
PythonFastAPINext.jsLangGraphGemmaElevenLabsMongoDBImentiv API
Hardening for launchAsk me about it: Clarity

Also shipped

GitHub

Skills

Languages

PythonDartTypeScriptC++JavaKotlinSQLJavaScriptC#

AI & Vision

PyTorchYOLOv8OpenCVComputational GeometryRAGGeminiGemmaLangGraphElevenLabsDINOv2FaissByteTrackTensorFlow

Engineering

FlutterNext.jsFirebaseCloud FunctionsDockerAWSGCPFastAPIPostgreSQLMongoDBVertex AIUnreal Engine

Education

Up next · Graduate school

M.S. in AI & Computer Science

Applying now · Fall 2027 start

Applications in progress

University of California, Irvine

B.S. Computer Science

Specialization in Intelligent Systems

Sept 2024 – Jun 2027 (expected) · Class of 2027

GPA 3.92
Dean's Honor List — All Quarters
Google Cloud Generative AI Leader

Monta Vista High School

High School Diploma

2020 – 2024 · Cupertino, CA

Contact

Open to internships, collaborations, or just trading notes on CV pipelines and RAG systems.

Open to SWE/ML internships — Summer 2027. Applying to MS programs for Fall 2027
Typical response within 24 hours.