Mohammed Ashraf
Bangalore
Summary
Innovative ML Engineer with 5+ years of experience designing AI-driven solutions. Specialized in Generative AI, LLMs, and scalable MLOps. Expertise in custom models for classification, NER, and intent recognition, with proven ability in fine-tuning Gemini 7B and LLaMA 2. Skilled in RAG systems for chatbots and knowledge bases. Proficient in Python, TensorFlow, PyTorch, FastAPI, and AWS. Experienced in leading teams and driving AI innovation in startup environments.
Overview
6
6
years of professional experience
1
1
Certification
Work History
Senior Machine Learning Engineer
Kapture CX
08.2023 - Current
- Engineered RAG-Based Multilingual Chabot With Semantic Chunking, Hybrid Search, And Multi-Vector Queries, Reducing Queries To Support Agents By 70%.
- Implemented Argentic RAG KMS, Boosting Support Agent Efficiency By 40% Through Graph-Based Knowledge Representation And Entity Disambiguation.
- Trained Custom Text Moderation Model With 97% F1 Score Using Transfer Learning.
- Developed NLQ Dashboard Using LLM, Enabling Natural Language Queries For Data Insights.
- Trained Custom T5-Based Intent & NER Models, Fine-Tuning Gemini 7B & LLaMA 2 For Customer Support.
Chief Technology Officer
Sales Sunday
07.2022 - 08.2023
- Developed And Deployed AI-Driven Sales Forecasting, Demand Planning, And Inventory Optimization, Yielding 25% Cost Savings.
- Implemented AI-Powered Auto QA System, Slashing Manual QA Costs By ~99% And Delivering Actionable Insights.
- Created Lead Profile-Based Script Generator, Boosting Conversion Chances By 80%, And Fine-Tuned LLM Models For Enhanced Performance.
- Developed RAG-Based KMS, Leveraging Knowledge From SOPs And Past Tickets To Provide Agents With Real-Time Answers And Insights.
- Spearheaded Automated Backend And Frontend System Development, Leveraging DevOps Pipeline For Seamless Deployment.
Lead Machine Learning Engineer
Wobb
09.2021 - 05.2022
- Developed Multimodal Classification Model Using CLIP, Combining Image And Text Data(Utilizing Word2Vec Embeddings), Categorizing Influencer Posts Across 37 Categories With 95.6% Accuracy.
- Implemented Advanced Influencer Discovery Algorithm Leveraging Social Media Profiles, Engagement Rates, Audience Demographics, And Graph-Based Clustering Techniques. Boosting Campaign Engagement By 90%.
- Designed Data Pipeline Using Apache Airflow, Gathering Data From Instagram, YouTube, Twitter, And LinkedIn, Preprocessing With NLTK, And Storing In AWS Data Lake For Influencer Profile Updates.
- Established MLOps Pipeline Using Kubernetes Optimizing Classification System Performance Through Automated Data Retrieval, Model Retraining, And Deployment.
Co-Founder
Frejourn
01.2019 - 06.2021
- Led development of robust and scalable backend infrastructure to support rapid growth.
- Implemented location-based algorithm and machine learning models for personalized user experience.
- Integrated third-party APIs to extend platform functionality and enhance user engagement.
- Successfully scaled platform to handle high traffic and ensured seamless user experience.
- Completed Y Combinator Startup School and Newchip Bootstrap Accelerator, gaining valuable mentorship and networking opportunities.
Education
Master of Science - Computer Science
REVA University
Bengaluru, Karnataka, India 08.2020
Bachelor of Computer Applications - Big Data Analytics
Jain University
Bengaluru, Karnataka, India06.2018
Skills
- Machine Learning
- Natural Language Processing
- Python
- FastAPI
- Generative AI
- Prompt Engineering
- LangChain
- HuggingFace Transformers
Certification
- Oracle Cloud Infrastructure 2024 Generative AI Certified Professional [Issued by Oracle] [08 Jul 2024]
- MTA: Introduction to Programming Using Python [Issued by Microsoft] [ 14 Nov 2019]
- API Designer Certification [API Academy] [12 Jan 2021]
- Data Analysis Using Python [Issued by IBM] [03 Apr 2020
PERSONAL PROJECTS
- Odia T5 Base Model For NLP Task In Odia Language: Developed a T5-based model for NLP tasks in the Odia language, including Question Answering (QA), summarization, and translation, using a custom Odia dataset. Trained on the mT5 architecture, achieving proficiency in various Odia NLP tasks.
- AgriQBot: Fine-tuned a FLAN-T5 model for question answering, using a custom dataset curated from trusted sources. Achieved a BLEU score of 0.86.
- Fine-Tuned Gemma 2B for Python Code Generation: Fine-tuned the Google Gemma 2B Instruct model with 4-bit quantization using PEFT. Employed Supervised Fine-Tuning (SFT) for accurate Python code generation from natural language instructions.
- Real-Time AI Telephony Sales Agent : Developed an AI-powered real-time agent for telephony sales with Twilio integration. The system responds to user queries in under 0.6 seconds. Integrated Groq and Deepgram for efficient speech recognition and NLP, enhancing lead conversion and sales efficiency.
Languages
English
Advanced (C1)
Hindi
Advanced (C1)
Timeline
Senior Machine Learning Engineer
Kapture CX
08.2023 - Current
Chief Technology Officer
Sales Sunday
07.2022 - 08.2023
Lead Machine Learning Engineer
Wobb
09.2021 - 05.2022
Co-Founder
Frejourn
01.2019 - 06.2021
Master of Science - Computer Science
REVA University
Bachelor of Computer Applications - Big Data Analytics
Jain University