.png){kind=logo}
Zufeshan Imran
Senior Machine Learning Engineer specializing in LLMs, RAG, and computer vision
Connect with Zufeshan
Zufeshan already has a relationship with Reval, so a warm intro from us gets a much better response than cold outreach.
Recommended
Already have an account?
About
Built an "AskMyVideo" system that turns YouTube videos into queryable knowledge graphs by transcribing audio (Whisper), chunking and embedding content, and enabling traceable answers back to exact timestamps. Strong in entity resolution (rules + fuzzy matching + TF-IDF/cosine with PR-curve thresholding) and modern retrieval stacks (FAISS, hybrid dense/sparse, domain fine-tuning with ~12% precision gain), with a production mindset using Airflow/Prefect, Docker/FastAPI, and LangSmith/Prometheus/Grafana observability.
Experience
Education
Key Strengths
- Built a system to chat with YouTube videos using knowledge graphs grounded in video content
- Designed a multimodal knowledge graph schema with time-aligned chunks and traceability to source timestamps
- Entity resolution pipeline combining rules + fuzzy matching + ML similarity scoring; achieved >95% match accuracy
- Threshold tuning using labeled validation set and precision-recall curves; added manual review bucket for borderline matches
- Improved search relevance via embeddings + FAISS and hybrid dense/sparse retrieval; ~12% precision lift from domain fine-tuning
- Production-grade Python workflow design with orchestration, validation, CI/CD, and observability/monitoring
- Clear prototype-to-production criteria using retrieval accuracy (>90%) and latency (<4s) SLAs; productionized with Docker + FastAPI + Prefect
Discover more candidates like Zufeshan
Search across thousands of pre-screened, high-quality, high-intent candidates on Reval.
Search Talent