Loading the portfolio

Software & AI Engineer

Aaryan Gajula

Building intelligent systems. Inspired by what comes next.

Explore my work

NowBuilding Te Vārtā · RemiVoice research · FIU, May 2027 Updated Oct 2026

  1. Mumbai, INOrigin

    Started undergrad at DJ Sanghvi College of Engineering ↗︎, dropped out after first year and moved to the US to be inside the AI revolution rather than read about it from across the world.

  2. Auburn, ALNov 2023

    At RFID Labs, Auburn University ↗︎, built data pipelines for 50,000+ RFID entries and a Neo4j schema mapping device dependencies.

  3. Miami, FLResearch

    In undergraduate research with Dr. Chen Chen ↗︎ at FIU, designed and built RemiVoice, a voice-first dual-agent system for reminiscence therapy with older adults living with mild dementia. Published as a first-author paper on arXiv ↗︎ in August 2026.

  4. Miami, FLIndustry · 2026

    Software Engineering Intern at CoOrdio Health (formerly OpenBots AI) ↗︎ from May to August 2026, building multi-tenant healthcare systems: cross-product recovery, form libraries, e-signatures, and deterministic + AI form processing. Earlier work included HIPAA-compliant agents and ETL pipelines for LTAC hospitals, compressing admissions and clinical review from days to hours.

  5. Iron Man got me into building. That part hasn’t changed.

How I build

Deterministic where it matters, generative where it delights.

Predictable cores, with models at the edges. Projects that show it are marked below.

1,527
GitHub contributions, past year (Oct 2026)
50+
developers guided through open-source contributions
Hours
for clinical review, down from days

REZTDemo · simulated

Claude Code five-hour window resets in 4:59:59

Software Engineering Intern · CoOrdio Health (formerly OpenBots AI)

May 2026 — August 2026

Miami, FL

  • Engineered a cross-product recycle-bin architecture in C#, ASP.NET Core, EF Core, and SQL for 18 entity types across a multi-tenant healthcare platform, implementing global query filters, tenant-scoped restore/delete operations, cascade recovery, filtered unique indexes, and configurable retention.
  • Designed backend architecture for a healthcare Form Library and e-signature platform spanning 19 tenant-scoped entities, packet assembly, role-based signing, audit trails, PDF generation, and 10 EF Core migrations.
  • Built a production deterministic + AI form-processing pipeline using Scriban, Azure OpenAI, field-level provenance, strict validation, and concurrency-safe background workers with atomic claiming, retries, and startup recovery.
  • Authored 220 unit/integration tests and 14 EF Core migrations, including a zero-data-loss redesign of the signing authorization model from a role hierarchy to independent ranked authority chains.
C# ASP.NET Core EF Core SQL Scriban Azure OpenAI Python .NET LangChain FastAPI HIPAA RPA
Additional work

Built HIPAA-compliant AI systems for LTAC hospitals, optimized production ETL pipelines in Python and .NET, developed LangChain + FastAPI agents for multi-step clinical review, and designed RPA workflows integrated with hospital data systems to compress days of manual processing into hours.

Founder & CEO · Te Vārtā

Present
  • Building a personalized news intelligence engine.
Supabase Postgres FastAPI Claude API Knowledge Graph

Founder & CEO · REZT

Present
  • Building REZT, a macOS companion and hosted dashboard that tracks Claude Code, Codex and Antigravity usage limits and alerts you on Telegram and Mac the moment they reset.
  • Stores only percentages and reset times, never tokens.
Python SwiftUI SQLite Telegram QStash

Research Assistant · Dr. Chen Chen, Florida International University

Present · RemiVoice, August 2026

Miami, FL

RemiVoice paper, arXiv 2026 (first author) ↗︎

  • Designed and built a dual-agent (conversational + image-generation) memory-recall prototype for dementia patients using Flask, the Gemini API, and React; coordinated focus group sessions with collaborators.
  • Researching conversation-aware AI agents and designing a memory architecture for dementia memory recall. Ember is the front-end testbed for this work.
Flask Gemini API React Memory Graph

Technical Program Assistant · Tech Talent Academy, FIU

Jan. 2026 — August 2026

Miami, FL

  • Managed the open-source contribution lifecycle for 50+ developers, partnering with BoxLang engineers to source and triage live issues and route contributors toward production-ready bug fixes and feature implementations.
  • Led students through Git-based open-source workflows, codebase navigation, debugging, and testing, helping them break down unfamiliar issues and carry fixes from local reproduction through review-ready submissions.
Git GitHub Actions CI/CD BoxLang SDLC
Additional work

Supported CIS 3590 through industry-standard Git workflows and review practices as a peer advisor.

Software Engineer (Mixed Reality) · INIT FIU

Sept. 2025 — Dec. 2025

Miami, FL

  • Architected a distributed networked state machine in Unity with C# using the Meta SDK to synchronize real-time multiplayer game state across devices, achieving 50ms end-to-end latency.
  • Reduced rendering overhead 20% on passthrough-compatible interfaces through URP pipeline optimization and Canvas Renderer tuning.
Unity C# Meta SDK URP

Software & Data Engineer · RFID Labs, Auburn University

Nov. 2023 — April 2024

Auburn, AL

  • Optimized automated data processing pipelines for 50,000+ RFID database entries using Python, SQL, and Kafka, reducing discrepancies by 15% and improving real-time tracking reliability.
  • Designed a Neo4j graph database schema modeling device relationships and dependency mappings, enabling 30% faster fault isolation and performance benchmarking.
Python SQL Kafka Neo4j CI/CD

  1. Founder · Product

    News that finds you. A personalized news intelligence engine.

    • Surfaces news that affects the people, companies, places and systems a user cares about, using a shared evidence-backed world graph plus a small per-user overlay.
    • FastAPI backend and native iOS app. The LLM sits only at the surface, so every story traces back to graph evidence.

    LLM at the surface only; the graph is the source of truth.

    Neo4j · FastAPI · Supabase · Claude API · iOS

    View project

  2. Product · Usage Reset

    Know the instant your Claude Code usage resets.

    • macOS companion plus a hosted dashboard that tracks Claude Code, Codex and Antigravity limits and notifies you via Telegram and Mac at reset.
    • Only percentages and reset times are stored, never tokens.

    Python · SwiftUI · SQLite · Telegram · QStash

    View project

  3. Voice agent

    A voice agent that turns customer calls into verified issues.

    • ElevenLabs + Twilio agent. A second-model check rejects unsupported claims and requires the same issue across 2+ calls before it counts as recurring.
    • Transcripts are treated as untrusted input: unanswered calls, prompt injection, grounding failures and malformed output are all tested.

    A second model rejects claims the transcript can’t back up.

    TypeScript · Next.js · ElevenLabs · Twilio

  4. Local AI · macOS

    Checks your work against your spoken goals, privately.

    • OCRs the focused window, deletes the screenshot immediately, and classifies work with a local Gemma 3 12B model.
    • Switched to enforced JSON schemas after finding an on-screen prompt-injection hole.

    Enforced JSON schemas instead of prompt-based parsing.

    Python · PyObjC · Swift · Apple Vision · Ollama

  5. AR coaching

    Real-time coaching through Meta Ray-Bans.

    • Sends first-person camera frames to Gemini 2.5 Flash every 3s and queues audio so guidance never overlaps.
    • Analyzes expert recordings and replays them as step-by-step audio.

    Python · FastAPI · Gemini 2.5 Flash · Swift

    View project

  6. Undergraduate research

    Voice-first AI for reminiscence therapy with older adults living with mild dementia.

    • Dual-agent design (conversational + image generation) built with Flask, the Gemini API and React. Advisor: Dr. Chen Chen.
    • Coordinated focus-group sessions with collaborators.

    Flask · Gemini API · React

    Read the paper

  1. Hackathon · San Francisco · Nebius, NVIDIA, Antioch, Toloka

    One challenge: build a Perception → Reasoning → Action loop. We had a working demo in about five hours, then pitched what comes next: a robot you teach just by talking to it.

    • A recursive feedback edge in the loop: a live critique signal feeds back into reasoning, so every attempt improves the next one in the same session, with no retraining.
    • Meta glasses as the interface: the robot sees what you see, you say what you want, and you correct it by voice. Hands-free, no code.

    Knowing which 5% to build so people can see the other 95%.

    Event

  2. Hackathon · Systematic Trading track

    Backfill, from FDA drug-shortage notices to auditable supplier-trading research.

    • Reconstructs historical shortage notices and supplier availability from Internet Archive captures, and maps each supplier to its listed owner on that date.
    • Tests whether manufacturers that can still supply a drug outperform their market hedge, with costs, controls and limitations reported openly.

  3. Academic honor

    Ranked top of the state in the 10th grade board finals.

    Roll of honor

  4. Cohort

    Technical AI Safety cohort, 2026.

    BlueDot

  5. Hackathon · Education track

    Shadow: real-time coaching through Meta Ray-Bans.

    • Sends first-person camera frames to Gemini 2.5 Flash every 3s and queues audio so guidance never overlaps.
    Devpost

  6. Certification

    Anthropic API certification.

    Verify

Florida International University (KFSCIS) ↗︎

Miami, FL

B.S. in Computer Science · May 2027 · GPA 3.51/4.0

Technical skills

Languages
Python, C#, TypeScript, Java, SQL, Swift
Frameworks
ASP.NET Core, .NET, EF Core, FastAPI, Flask, React, Next.js, Unity, PyObjC
AI & ML
Azure OpenAI, Gemini API, Gemma, Ollama, ElevenLabs, Apple Vision OCR
Cloud & Tools
Azure, Docker, Kafka, Git, Twilio
Databases
PostgreSQL, Azure SQL, Neo4j

↑↓ select↵ goesc closeanswers by OpenAI