Looking for a job

Anahit Manukyan

@anahitmanukyan

I build speech-to-text, RAG, and ML pipelines that turn data into useful insights.

I've built speech-to-text and RAG pipelines at Spesna, processing more than 50 hours of audio with OpenAI Whisper and improving internal knowledge-base retrieval latency by 30%.

I design prompt engineering workflows for structured LLM outputs and prepare speech datasets for fine-tuning through audio chunking and text alignment.

At Picsart Academy, I scraped and modeled Armenian real estate data, building XGBoost and LightGBM price-prediction models that achieved an R² of 0.92 on held-out data.

Through personal projects, I've built an EPL semantic-search RAG system over 2,000+ matches and an e-commerce analytics dashboard that helped identify revenue drivers and improve inventory turnover by 10%.

Experience

Work history, roles, and key accomplishments

Current

AI/ML Intern

Current

Spesna

May 2026 - Present (4 months)

Built a speech-to-text pipeline with OpenAI Whisper, processing 50+ hours of audio data. Developed a RAG system using LangChain and FAISS that reduced information retrieval latency by 30% and improved response accuracy for internal knowledge base queries.

OpenAI Whisper Lang Chain Faiss RAG Prompt Engineering Speech to Text LLM Fine tuning Audio Processing

Data Science Intern

Picsart Academy

Feb 2026 - Apr 2026 (2 months)

Scraped and cleaned housing data from Armenian real estate websites using Python and pandas. Built and compared ensemble models (XGBoost, LightGBM) to predict apartment prices, achieving an R² of 0.92 on held-out data.