Promila Ghosh

AI Specialist

🌍 Khulna, Bangladesh

Let's Connect!

LinkedIn GitHub Google Scholar

Hugging Face Medium X(Twitter)

Learn more about me!

Experience Projects Education Publications Technical Skills Conference Presentations Participation

Experience

Data Scientist - United We Care (July 2022 - March 2025)

Built the United-MedASR system for medical transcription, cutting WER to 5.96% and boosting speed 7–10x.
Generated 5,486+ hours of synthetic clinical audio using Diffusion Models and GPT — open-sourced on Hugging Face.
Built a multimodal agent using LangGraph and LMMs to analyze medical documents (PDFs, images, handwriting) with ~70% accuracy.
Developed facial and gesture analysis systems using Redis and sliding windows — enabling faster real-time processing (~60% accuracy).
Designed a real-time data pipeline to stream user conversations from S3 to BigQuery, powering analytics for an app with 10K+ downloads.

Founder & AI Research Scientist - The Data Dilemma (May 2025 – Present)

Open-sourced MediBeng, a Bengali-English medical speech dataset with 4K+ samples.
Fine-tuned MediBeng Whisper Tiny, achieving 0.98 BLEU (vs 0.3 baseline) for clinical translation.
Developed Medibeng-Orpheus-3b for healthcare speech translation with code-switch and privacy support.
Built a real-time chat interface using WebSocket, FastAPI, and Groq API for seamless LLM output.
Engineered a healthcare RAG system with hierarchical context and ChromaDB for accurate retrieval.

Technical Content Writer - Qtec Solution Limited (March 2021 - November 2021)

Authored 20+ programming courses on Machine Learning and Data Structures.
Worked with designers and developers to ensure clarity, quality, and depth.

Projects

MediBeng Whisper Tiny: Improves doctor-patient transcription by training the Whisper Tiny model to translate mixed Bengali-English speech into English for better analysis and record-keeping. Link
ParquetToHuggingFace: Processes raw audio data, converts it into Parquet files, and uploads them to Hugging Face for easy access and model training. Link
GroqStreamChain: A real-time AI-powered chat application that streams AI responses via FastAPI, WebSocket, and Groq for interactive, low-latency communication. Link
Medical Document Processing – United We Care: Developed an intelligent agent that reads and classifies medical documents (e.g., prescriptions, health summaries) using OCR and image processing for better mental health advice. Link

Education

M.Sc. Engg., Computer Science and Engineering - Khulna University (2022 - 2023)
B.Sc., Computer Science and Engineering - North Western University (2017 - 2021)

Publications

Computer Vision and Speech Technology

Recognition of Sunflower Diseases Using Hybrid Deep Learning and Its Explainability with AI, in Mathematics, vol. 11, no. 10, pp. 2241, May 2023. Link
MediBeng Whisper Tiny: A Fine-Tuned Code-Switched Bengali-English Translator for Clinical Applications, in medRxiv, April 25, 2025. Link

Natural Language Processing (NLP)

Multi-labelled Bengali Public Comments Sentiment Analysis with Bidirectional Recurrent Neural Networks (Bi-RNNs), in Applied Intelligence for Industry 4.0, May 2023, DOI: 10.1201/9781003256083. [Online]. Available: Link
Fake News Detection of COVID-19 Using Machine Learning Techniques, in COVID-19: A Machine Learning Perspective, 2022, DOI: 10.1007/978-3-030-93247-3_46. [Online]. Available: Link
Safeguard: A Prototype of An Application Programming Interface to Save the Disaster Affected People, in Proceedings of the 10th International Conference on Computing, Communication and Networking Technologies (ICCCNT), 2019, DOI: 10.1109/icccnt45670.2019.8944883. [Online]. Available: Link

Healthcare

A Comprehensive Analysis on Risk Prediction of Acute Coronary Syndrome Using Machine Learning Approaches, in Proceedings of the 21st International Conference of Computer and Information Technology (ICCIT), 2018, DOI: 10.1109/iccitechn.2018.8631930. [Online]. Available: Link
Risk Prediction of Ischemic Heart Disease Using Artificial Neural Network, in Proceedings of the International Conference on Electrical, Computer and Communication Engineering (ECCE), 2019, DOI: 10.1109/ecace.2019.8679362. [Online]. Available: Link
Typical and Non-Typical Diabetes Disease Prediction using Random Forest Algorithm, in Proceedings of the 11th International Conference on Computing, Communication and Networking Technologies (ICCCNT), 2020, DOI: 10.1109/icccnt49239.2020.9225430. [Online]. Available: Link
An Empirical Study on Diabetes Mellitus Prediction Using Apriori Algorithm, in Advances in Intelligent Systems and Computing, vol. 1328, pp. 481-493, 2021, DOI: 10.1007/978-981-15-5148-2_48. [Online]. Available: Link
Human Behavior Analysis using Association Rule Mining Techniques, in Proceedings of the 11th International Conference on Computing, Communication and Networking Technologies (ICCCNT), 2020, DOI: 10.1109/icccnt49239.2020.9225662. [Online]. Available: Link
A Machine Learning Approach to Identify the Correlation and Association among the Students' Educational Behavior, in Proceedings of the International Conference on Computing Advancements, 2020, DOI: 10.1145/3377049.3377130. [Online]. Available: Link

Technical Skills

Deep Learning: Neural Networks, CNNs, RNNs, Transformers (BERT, GPT)
Programming: Python (Pandas, NumPy, TensorFlow, PyTorch)
Speech Recognition: Whisper, Wav2vec, Speech-to-Text APIs
Cloud Computing: AWS (EC2, S3), GCP (Compute Engine, Google Cloud Storage)

Conference Presentations

International Conference on Big Data, IoT and Machine Learning (2021)
International Conference on Computing, Communication and Networking Technologies (2019)
21st International Conference on Computer and Information Technology (2018)

Participation

Intra University Programming Contest: Secured the championship (2019)
ACM ICPC Asia Dhaka Regional Contest: (2018)
Macroinnovators IEEE SS12: Qualified for primary selection (2018)