Abin Varghese
I love Coffee and building AI solutions that scale.
Co-Founder & CEO @ Coffee AI
Wanna know me? "Talk with Abin" in the bottom right!
Read my essays on AI research, engineering, & technical leadership.
Work Experience
A chronological record of academic and industry positions in AI engineering, machine learning research, and applied data science.
Lead AI Solutions Architect
Coffee AI • London, UK
Real-time voice AI platform that captures expert tacit knowledge, delivers it as live coaching, and benchmarks candidates against top performers.
Key Contributions
- Sub-200ms voice latency pipeline: WebSocket audio streaming with server-side VAD, ElevenLabs TTS, and gapless audio playback queue - reduced perceived response time by 60%
- 5-way RRF RAG retrieval system using pgvector with OpenAI text-embedding-3-large and Jina CLIP, enabling sub-second semantic search across 10K+ knowledge segments
- Speculative prefetch architecture: question detection heuristics triggering parallel RAG queries during user speech, reducing perceived response time by ~40%
- Edge-first privacy architecture (Go): daemon-based workflow capture with on-device keystroke monitoring (count-only privacy mode), 30fps video recording, NDJSON event aggregation at <5% CPU overhead
- Production infrastructure: microservices across Railway (Python FastAPI, Go) and Vercel (Next.js) with PostgreSQL connection pooling, Redis caching, automated CI/CD - 99.5% uptime
- AI candidate screening system using embedding similarity scoring to benchmark candidates against top performer response patterns
- Won £10,000 Kingston University Enterprise Circle Award (out of 100+ applicants)
Senior AI Engineer
Contentful • London, UK
CX-focused AI engineering at the composable content platform serving 30% of the Fortune 500.
Key Contributions
- CX
GenAI Solutions Architect
GIMO • London, UK
Built AI agents for compliance, fraud detection, translation, recruitment, and customer support across 12 markets.
Key Contributions
- Compliance AI Agent: Automated safer gambling reports and AML flagging - 90 minutes per user down to under a minute, 98% faster. Saves thousands of work hours yearly
- Risk, Fraud & Payment Agent: Automated fraud checks on new player registrations. Each compliance officer saves ~6 hours daily. ~89% accuracy
- Translation AI Agent: Built in-house translation beating DeepL for accuracy across 12 markets. Cut translation costs by £120,000/year
- AI Recruitment System: Conversational AI screening candidates via chat with vector scoring. Cut screening time by 85%
- Nexus AI Chatbot: Built with Parlant instead of buying a £50,000 enterprise solution. 67% NPS and improving
- EU AI Act Compliance Training: Trained 356 employees across UK, Romania, Malta, Italy. 99.1% completion, 8.19/9 assessment score, 4.8/5 satisfaction
Data Scientist
VIRAL • London, UK
Developed RAG-based systems and automated ETL pipelines for job descriptions analysis.
Key Contributions
- Developed a RAG-based ranking system using sentence_transformers embeddings and cosine similarity, evaluating job descriptions against 25 KPIs, resulting in a 40% increase in high-quality candidate applications and demonstrating advanced NLP skills
- Engineered an automated ETL pipeline using Python, and Airtable to extract job descriptions from multiple job boards, leveraging Axiom.ai, BeautifulSoup and Selenium for web scraping, reducing data collection time by 90% and increasing processed listings by 3x
- Developed comprehensive data cleaning and transformation processes using Pandas and NLTK, including regex pattern matching, text normalisation, and entity recognition, improving data quality by 95% and ensuring consistency across diverse job posting formats
Machine Learning Research Engineer (Computer Vision)
FruitCast • Lincoln, UK
Built custom Visual Odometry solutions for agricultural settings.
Key Contributions
- Built a custom Visual Odometry solution for polytunnel environments by adapting the LightGlue feature extractor. Optimised the algorithm to process laterally moving video footage, a significant departure from traditional forward-facing camera implementations
- Built a Visual Odometry pipeline, conducting on-site testing at the Riseholme campus by setting up cameras in polytunnels and processing monocular camera footage. Applied Gaussian blur to smooth extracted features in each frame, enabling accurate distance estimation
- Spent approximately 10% of my time on data annotation, classifying strawberries into six distinct categories using V7, ensuring high-quality training data for machine learning models
Data Scientist
Infosys • Bangalore, India
Led development teams, contributed to ML projects, and created educational courses.
Key Contributions
- Engineered team leadership: Orchestrated development of an enterprise solution utilising the Attention model to autonomously extract and categorise skills from resumes, leading to substantial cost savings in HR operations at Bangalore and Mysore DC
- Contributed to ML projects: Enabled BEF's global impact, attracted external clients like GlobalFoundries, boosting profits in 2021-2022
- Created courses: Trained 5000+ employees in ML, NLP, and stats, indirectly increasing revenue
- Awarded best author: GPT-3 course in Q2 2022, live to 250,000+ internal employees and external vendors
Leader for Data Science and Machine Learning
Corizo • Remote
Led educational initiatives and developed data science curriculum.
Key Contributions
- Achieved 55% revenue increase: Led 1000+ students in data science and machine learning, driving growth
- Headed highest-revenue department: Developed impactful lessons and projects
Intern
Infosys • Bangalore, India
Trained in various technologies and fundamentals of data processing.
Key Contributions
- Got Trained in IVS-DWT stream
- Worked on UNIX, PLSQL, Python Programming language, Machine Learning fundamentals, Big Data, BI Testing, and extensively worked on ETL using Informatica
Applied Privacy Podcast [archived]
A collection of interviews with industry experts, researchers, and thought leaders in the field of privacy technology and data protection.
Interview with Jeff Jockisch
Jeff, PrivacyPlan's CEO and Data Privacy Researcher, explores diverse privacy facets via podcasts, expert conversations, and data analysis. His focus areas include tech, data brokers, regulations, AI, and podcast evaluations.
Interview with Patricia Thaine
Patricia is the CEO and co-founder of Private AI, a company that she started with Pieter Luitjens, CTO of Private AI whose mission is to build the privacy layer for software.
Interview with David Burnett
Dave, Zero Biometrics' business development head, designs secure face authentication without storing sensitive biometrics, reducing data breach risks. Protecting biometric data, whether on your device or in the cloud, is crucial.
Interview with Robin Andruss
Robin is the chief privacy officer at Skyflow. Robin is a strategic risk & business-focused Privacy & Data Protection Leader with over 15 years combined in the privacy, risk, audit, finance, strategy, and compliance space.
Interview with Thomas La Piana
Thomas is the senior software engineer at Ethyca. Thomas and I talked about the privacy landscape from an engineering perspective, inserting privacy into the CI/CD pipeline, the future of privacy tech in the coming years and much more!
Interview with Tawfiq Alashoor
Dr. Tawfiq Alashoor, an Assistant Professor at Copenhagen Business School's Department of Digitalization, holds B.S., M.S., and Ph.D. degrees.
Contact Information
For collaboration opportunities, consulting inquiries, or professional networking, please use one of the communication channels below.
Professional Networks
Send a Direct Message
Please complete the form below for project inquiries, speaking engagements, or collaboration proposals.