I'm a student studying Data Science and Computer Science at UC Berkeley.
I'm currently at Medoh building high‑performance search and personalization systems to digitize and scale the doctor–patient experience.
I'm interested in applied NLP, LLM infrastructure, and clever, intuitive interfaces.
You can find my previous experience on Linkedin and my work on Github.
Reach out at aryamankukal [at] berkeley [dot] edu.
When I'm not building, you'll find me with my dog Astro, trying an ethnic food spot, writing, at a concert, in the gym, or on a late‑night drive blasting R&B.
-
🛠️ Currently working on:
- Medoh Health → Building an LLM-powered semantic search system (embeddings, pgvector, caching, personalization)
- FHL Vive Center @ UC Berkeley → Research assistant on TAI Team, developing an LLM-powered educational agent for Berkeley courses (RAG + orchestration)
- Dhisana AI → Contributing to GTM agent infrastructure as an open-source developer (Dockerized workflows, LangChain/LangGraph)
-
📚 Learning: Semantic search, LLM optimization, vector DBs (FAISS/Chroma), LangChain, and agent orchestration
-
🤝 Open to Collaborate: Looking to contribute to startups or research in NLP, GenAI infra, or agentic systems
-
💬 Ask me about: LLMs, backend infra, hackathon strategy, vector search, open-source tools
-
⚡ Fun Fact: I love singing (amateur R&B enthusiast), vibing to music, and editing videos — back in the day, I ran a YouTube channel with skits, short films, and trickshot videos.
- Languages: Python, Java, JavaScript, SQL, HTML/CSS
- ML/NLP: Scikit-learn, TensorFlow, Keras, PyTorch, HuggingFace, NLTK, VADER, PyTesseract
- Frameworks/Infra: Flask, FastAPI, Node.js, React, Next.js, AWS EC2, Docker, Electron.js, Tailwind, Git
- Other Tools: Chroma, Groq, Vapi.ai, Gemini API, Manim, SQLite, Jupyter, Pandas, NumPy, LaTeX
- Certifications:
-
🎓 Scholara.ai – Explains academic papers using LLMs, citation graphs, and voice agents
GitHub | Demo -
🛡️ ChatGuard – Detects and classifies harassment using NLP + OCR
Demo | 🏆 1st Place – Congressional App Challenge -
🧠 AudioLec – Lecture transcriber that extracts key topics & suggests videos
🏆 2× Hackathon Winner | TSA Nationals Top 10
-
🧑💻 Software Engineer @ Medoh
Built an LLM-powered semantic search system using Mistral/BGE embeddings with pgvector; automated embedding updates via PostgreSQL triggers, added caching & personalization layers, and improved precision from ~30% → ~95%. -
🧑💻 Software Engineer @ Playdo.ai
Built GPT-4 tools to validate test cases and redesigned backend with Flask, Node.js & AWS — served 150+ users with <200ms latency -
🌍 Open Source @ Dhisana AI
Contributing to go-to-market AI tools and agent pipelines for GTM automation -
🧪 ML Researcher @ Cambridge Centre for International Research
Built PTSD/GAD dataset, applied PCA + GMM for label refinement, used for semantic model training -
🛠️ Intern @ Tech For Good Inc
Helped design Assurance — a real-time ML-based weapon detection system for schools -
🎓 Founder & Exec Roles
Codefy, SparkCS, EduVantage, The Academically Driven — ran tech education & nonprofit initiatives, built learning platforms
Before building LLM agents and scaling ML infra, I was just a kid experimenting with Python on my family laptop. Some of the first programs I ever wrote (ages 10–12):
-
🕰️ Live Virtual Analog Clock
A real-time analog clock built with Python’s turtle and datetime, featuring synced hands, smooth animations, and custom-drawn visuals—no external libraries used. -
🧮 Calco
A voice-controlled calculator built with Python that listens to spoken math problems using speech_recognition, parses the input, performs the operation, and speaks the answer aloud -
🎥 Cinema Movie Program
A simple yet functional movie ticketing system with age restrictions, real-time seat tracking, and user interaction, all built with basic Python logic.
- 📧 [email protected]
- 🌐 LinkedIn | GitHub | Devpost
Thanks for stopping by! 🚀