lindita.exe

Lindita Kalaj

Software Engineer · AI & Backend · New York

Work log

April 2026 to now

Built the newsletter platform in Next.js, TypeScript, Node.js, and Postgres/Drizzle. One codebase serves multiple regions. The largest has 1,100 subscribers at a 65% open rate.
Wrote the automated content pipeline, from web search for local sources through drafting with Sonnet 4.6 to delivery, with schema validation and automatic retries on malformed model output.
Shipped a self-serve ad portal on Stripe where advertisers buy placements directly. An LLM step matches each ad to the most relevant section of the next issue and inserts it automatically.

April 2025 to February 2026

Built a real-time transcription prototype for 60 minute meetings using Whisper and Llama 3 with token chunking and speaker diarization, reaching 95% transcription accuracy in QA comparisons.
Prototyped a headless browser agent (Puppeteer/Selenium) that joins scheduled Google Meets in under 30 seconds, identifies active speakers, and records audio for downstream transcription.
Prototyped a RAG system over thousands of documents using OpenAI embeddings and OpenSearch vector indexes, retrieving relevant passages at query time to build LLM context.

June 2023 to February 2025

Trained a YOLO-based screen detection model that identifies on-screen UI elements, automating data retrieval from devices so results stayed current without manual checks.
Built an autonomous agent that drives HDMI devices through multi-screen UI flows using YOLO for perception, adapting to content that changed daily without handwritten scripts.
Developed and maintained a custom Selenium-based library wrapper to streamline data scraping from streaming platforms, reducing client data retrieval time by 80%.

January 2021 to May 2022

Split Check · Android app that splits a restaurant bill by item from a photo of the receipt. Java, ML Kit, SQLite.
Clippy Desktop Pet · A little Clippy that sits on your desktop and answers questions. Python, PyQt6, OpenAI API.
Vector Remote Control · GUI for driving the Vector robot over its live camera feed. Python, Tkinter, Vector SDK.

AI / agents: Anthropic API, OpenAI API, Gemini, Llama, function calling / tool use, autonomous agents, RAG, embeddings, vector databases, prompt engineering, LangChain, Ollama
Languages: Python, TypeScript, JavaScript, Java, HTML/CSS, Bash
Frontend / backend: React, Next.js, Node.js, REST APIs
Automation: Puppeteer, Selenium, YOLO
Databases: PostgreSQL, Drizzle, MySQL, MongoDB, SQLite, OpenSearch
Cloud / DevOps: AWS, Docker, Kubernetes, CI/CD, Git

B.A. Computer Science, CUNY Queens College, September 2022. Dean's List, GPA 3.83.

Crocheting, drawing, nature walks, traveling.