Full-Stack & AI/ML Software Engineer · 3+ years shipping production web & mobile at scale

🎓 MS in Computer Science — UT Dallas (Fall 2026, Intelligent Systems) · 🔬 Generative AI Systems Research Assistant — UTD ATC 3D Studio (LightSquad) · 📍 Richardson, TX

🟢 Open to SWE & AI/ML internships and new-grad roles — I love building reliable systems end to end.

I build production-grade software from the ground up — from cross-platform apps used by 100K+ people to local-first, open-source AI tooling. Previously an SDE at Reliance Jio and Product Engineer at Yosemite Crew; now focused on AI/ML systems and engineering open source to a professional bar (typed end-to-end, tested, CI-gated, documented).

ATLAS — a multi-team AI agent system built like infrastructure, not a demo: an orchestrator routes goals to specialist teams, a Critic verifies every result before it counts as done, three memory tiers (episodic SQLite ledger · semantic store · versioned skill playbooks) make it compound over time, and destructive host actions are human-approval-gated in code, not prompts. Three interchangeable agent runtimes (Claude Agent SDK / LangGraph / fully-local Ollama) behind one typed protocol, a live WebSocket dashboard, an eval harness gating CI, and ready-to-load ComfyUI / TouchDesigner / StreamDiffusion workflows for generative-video work on a 10 GB VRAM budget. Live dashboard →

Python · FastAPI · asyncio · Next.js 14 · WebSocket · SQLite · mypy --strict · 55 tests incl. full e2e · behavioral evals in CI · Storybook · Vercel CD · MIT

Job Sentinel — a local-first, open-source career platform: searches jobs across multiple sources, scores how well your profile matches a role with a local (or bring-your-own-key) LLM, tracks your whole application pipeline, and generates ATS-tuned résumés & cover letters — all on your own machine, nothing leaves it. Ships with a clip-to-track browser extension, a Next.js web UI, a Typer CLI, and a FastAPI backend. Live demo → · docs

Python · FastAPI · Playwright · Next.js/React · SQLite · mypy --strict · ~450 tests · CI gates: CodeQL · OpenSSF Scorecard · gitleaks · pip-audit · license compliance · reproducible uv.lock builds · MIT

Kalki — an autonomous AI content-generation platform: LLM-scripted short-form video with automated media processing and scheduled publishing.

Generative AI Systems Research Assistant — UT Dallas, ATC 3D Studio (LightSquad) · 2026– Real-time generative pipelines on constrained GPUs: StreamDiffusion deployment on Windows/CUDA (xformers + TensorRT acceleration, benchmarked), an OSC-controlled TouchDesigner bridge for live AI video, and ComfyUI workflows for LoRA training and Wan 2.2 image-to-video inside a 10 GB VRAM budget.

Product Engineer — Yosemite Crew (DuneXploration UG, Germany) · Remote · 2025–2026 Single-handedly built the entire cross-platform React Native app and the web PIMS frontend for an open-source veterinary platform that grew to 2.1k+ GitHub stars, 78 forks, 3,100+ commits. Integrated IDEXX Labs & Merck Veterinary Manual APIs; ran the full AWS stack (Amplify, Cognito, S3, CloudFront) plus Supabase/Redis; shipped to the App Store & Play Store; authored 14 GitHub Actions workflows (CodeQL, Gitleaks, SonarCloud, Chromatic, Playwright E2E). (Full-time under the company account @harshit-yc.)

Software Development Engineer I — Reliance Jio Platforms · 2023–2025 Built and maintained high-traffic web & mobile apps serving 100K+ monthly active users (React, Next.js, Angular, React Native). Shipped a shared React + TypeScript component library adopted by 3+ teams and cut average page load ~40% via SSR and performance tuning.

Agentic systems — ATLAS: multi-team orchestration with critic verification, three-tier memory, and approval-gated host execution; provider-agnostic across Claude Agent SDK, LangGraph, and fully-local Ollama.

Real-time diffusion — StreamDiffusion on consumer CUDA hardware (xformers/TensorRT, benchmarked), TouchDesigner live-video bridge over OSC, ComfyUI LoRA-training and Wan 2.2 video workflows engineered for a 10 GB VRAM ceiling.

Parkify — Automated Number-Plate Recognition (undergraduate research): an end-to-end YOLOv5 + custom-CNN pipeline reaching 95.45% mAP detection and 89.66% character-recognition accuracy, with an OpenCV preprocessing stack and a local CUDA training setup.

Languages: TypeScript · Python · JavaScript · Java Frontend/Mobile: React · Next.js · Angular · React Native · Tailwind Backend/Data: Node/Express · FastAPI · Spring Boot · Supabase · Redis · MySQL · MongoDB Cloud/DevOps: AWS · Docker · CI/CD · SonarCloud · Playwright AI/ML: PyTorch · OpenCV · YOLOv5/CNNs · agentic systems (Claude Agent SDK · LangGraph · Ollama) · RAG · diffusion pipelines (StreamDiffusion · ComfyUI · Wan 2.2) · CUDA/TensorRT · quantization

• M.S. Computer Science, The University of Texas at Dallas — Intelligent Systems, Fall 2026
• B.E. Computer Engineering, University of Mumbai (Vidyalankar Institute) — CGPA 9.53/10