Skip to content

AI & full-stack engineer

I build AI systems that hold up beyond the demo.

LLM infrastructure, durable workflows and applied ML—built end to end.

Explore my work
View résumé (PDF, opens in a new tab)
Portrait of Vaibhav Gupta
Vaibhav GuptaAI Engineering Intern · Oliv AI
RequestGatewayretry · breakerOpenAIprimaryGeminifallback
Interactive simulationhealthy

Primary provider serving

The request goes through the gateway to OpenAI, and tokens stream straight back.

Illustrative simulation of InferLens’ routing design—not a live trace. How InferLens does this

01Selected work

Systems built to be inspected, interrupted and trusted.

Three projects I'd point an engineer to first, each with real interfaces, honest status labels and a case study covering how it fails.

InferLens landing page: 'See what happens behind every LLM response' with a request signal panel

Request path

  1. ConsoleNext.js
  2. GatewayFastAPI
  3. ProviderOpenAI · Gemini
  4. StreamRedis Streams
  5. Workeridempotent
  6. StorePostgreSQL
Live demoLLM infrastructure

InferLens

See what every LLM request costs.

A provider-agnostic LLM gateway with streaming, fallback and traces for comparing latency, tokens and estimated cost.

  • SSE streaming, cancellation, bounded retries, provider fallback and circuit-breaker recovery
  • Event-driven logging through Redis Streams, background workers and dead-letter recovery
  • Console for traces, replay and side-by-side provider comparisons
Next.jsTypeScriptFastAPIRedis StreamsPostgreSQLPrometheus
Live demo for InferLens (opens in a new tab)Code for InferLens (opens in a new tab)
LectureWeave landing page: 'Turn every lecture into notes you can trust' with a transcript and cited note preview
Live demo

LectureWeave

From live lecture to source-grounded notes.

A learning workspace that connects live transcripts, cited answers, quizzes and flashcards through hybrid retrieval.

  • Streams lecture audio over WebSockets and transcribes it live with Whisper
  • Hybrid retrieval: Atlas Vector Search plus keyword search, merged with Reciprocal Rank Fusion
  • Clickable citations, subject-level chat, quizzes and flashcards
ReactFastAPIWebSocketsWhisperMongoDB AtlasRRF
Live demo for LectureWeave (opens in a new tab)Code for LectureWeave (opens in a new tab)
OrderPilot control room for order ORD-4K2P9Z awaiting approval, with two customer messages waiting on a human decision
Source only

OrderPilot

Workflows that survive interruptions.

An order supervision system with durable execution, human approvals and recovery across failures.

  • One durable Temporal workflow per order, driven by signals and durable timers
  • Sensitive actions wait for a human to approve or reject them
  • 96 automated tests covering worker restarts, database failures and continuations
TemporalFastAPINext.jsPostgreSQL
Code for OrderPilot (opens in a new tab)

Also built

AlertSphere admin dashboard with an alert form for title, message, severity, delivery type, reminder frequency and visibility
Source only

AlertSphere

Alerts that reach the right people, on schedule.

An organisational alert and reminder platform: targeted alerts, severity levels, scheduled reminders, per-user read and snooze state, and team analytics.

FastAPI · SQLAlchemy · APScheduler · React · Recharts

Code for AlertSphere (opens in a new tab)
Video walkthrough

LinkedIn Profile Classifier

A webhook pipeline that verifies before it trusts.

Classifies LinkedIn profiles by team and seniority with n8n and Apps Script, behind signed, replay-protected webhooks.

n8n · Google Apps Script · Webhooks · HMAC-SHA256 · Apify

Watch walkthrough for LinkedIn Profile Classifier (opens in a new tab)

02Independent research · Applied ML

BugClassiNet. What kind of bug is this?

Bug trackers fill up with reports that look alike but behave very differently. BugClassiNet is a three-step research pipeline that first decides whether a report describes a bug, then whether it fails deterministically, and finally whether it is an aging-related bug that worsens the longer a system runs. The work also measures how a decision early in the pipeline carries through to the final answer.

≈1.09M

cleaned training reports

approximately

142,320

official test reports

first step

0.7934

Macro-F1, first step

issue-type classification only

0.7934 is the first step's reported result—not the accuracy of the whole pipeline. Evaluated separately on 806 reports already known to be bugs, the complete pipeline reported 0.6799 accuracy; that does not measure arbitrary incoming reports.

Results are as reported by the research, not independently reproduced here. Detailed reports and unpublished methods are not part of this site.

03Experience

Production workflows, real customers, real constraints.

  1. Jan 2026 – Present

    Remote

    AI Engineering Intern · Oliv AI

    • Built and maintained production API-driven workflows using n8n, REST APIs, webhooks, schedulers, retries, validation and error handling for reliable asynchronous execution.
    • Integrated HubSpot, Salesforce, Clay and external services through API-based automation, implementing data routing, synchronisation and customer-specific business logic.
    More detail
    • Developed internal AI workflow systems with structured prompts, data models, validation logic and feedback mechanisms to improve reliability and task execution.

  2. Jun 2025 – Aug 2025

    Remote

    Product Analyst Intern · Wealthy Nivesh

    • Processed and analysed financial, portfolio and operational datasets using Python and Pandas to identify inconsistencies and support reporting and decision-making.
    • Documented workflows, business rules and edge cases across product, operations and compliance processes.

Student venture

Aug 2024 – Present

VzenCare (VisionV-Soft)

Co-Founder · Student startup

  • Led technical development of an AI/ML retinal disease detection system using retinal image datasets for early-stage screening.
  • Applied machine learning and computer vision models in Python for classification and pattern recognition.

VisionV-Soft placed first nationally at StartupThon 2025.

Education

2023 – 2027

Thapar Institute of Engineering and Technology

Bachelor of Engineering, Computer Science Engineering

Patiala, IndiaCGPA 8.1

Portrait of Vaibhav Gupta

Patiala, Punjab, India

04About

I like the parts of a system nobody demos.

I'm a computer science undergraduate at Thapar Institute and an AI Engineering Intern at Oliv AI, where I build the production workflows that connect language models, CRMs and external APIs.

Outside work I build systems end to end—gateways, durable workflows, retrieval pipelines—and study where machine learning models break when they leave the benchmark.

  • Make it inspectable

    Every request should leave a trace you can read later.

    See InferLens →

  • Plan for the failure

    Assume the worker restarts and the provider times out—then design for it.

    See OrderPilot →

  • Measure honestly

    Report the ceiling, the real number and the gap between them.

    See BugClassiNet →

Tools, and where I’ve used them

Hover, focus or tap a logo to see where it’s used.

AI & Retrieval

Backend & Workflows

Frontend & Interfaces

Data & Infrastructure

05Recognition

Placements, finals and service.

Each entry links to the public post or certificate that records it.

Community & coordination

Helping run international hackathons.

Round II coordination for online AI hackathons, recognised with certificates of appreciation.

  • Session Coordinator, Round II

    Cyber AI Hackathon 2026

    University of Derby, UK · technically supported by the American Society for Engineers

    Coordinated Round II sessions of the online hackathon, held 10–11 October 2026.

    ID CyberAI2026/SessionChair/017Certificate (PDF) (opens in a new tab)

  • Coordinator & Judge, Round II

    AI for Sustainability Hackathon 2026

    Canadian University Dubai, UAE · technically supported by the American Society for Engineers

    Coordinated Round II and judged participant projects on AI-driven sustainability, 14–16 February 2026.

    ID SustainAI2026/Coordinator/020

  • Session Coordinator, Round II

    Cyber AI Hackathon 2025

    University of Derby, UK · technically supported by the American Society for Engineers

    Coordinated Round II sessions and supported structured assessment of submissions, 21–22 November 2025.

    ID CyberAI2025/SC/008Certificate (PDF) (opens in a new tab)

06Contact

Have a system worth building?

For engineering opportunities, collaborations, or a conversation about the work.

Email

guptavaibhav2708@gmail.com

Open in Gmail (opens in a new tab)Email app

Phone

+91 70499 15514

Call

Elsewhere

GitHub (opens in a new tab)LinkedIn (opens in a new tab)Résumé (PDF, opens in a new tab)

LinkedIn not loading? Try the alternative link (opens in a new tab)

Or send a message

Fields marked * are required.