Arjit Bhardwaj
AI at the core. First principles applied wherever curiosity goes next.
I believe AI will become as fundamental to society as electricity, not because it replaces human judgment, but because it amplifies it. The most meaningful AI systems won't simply generate answers; they'll enable better thinking, better decisions, and better outcomes. That belief drives everything I build, from trustworthy financial intelligence today to technologies that could one day contribute to humanity's next great frontier.
That mindset has shaped my journey, from improving autonomous navigation in adverse weather at IIT Hyderabad, to developing deep learning solutions for agriculture at ISRO, to building trustworthy financial intelligence at Bharosa. Whether it's wealth management, open-source projects, or autonomous systems for lunar navigation, I've always been drawn to solving problems where technology can create lasting, meaningful impact.
I enjoy building at the intersection of artificial intelligence, software engineering, and product thinking. I'm particularly interested in agentic AI, reasoning systems, financial intelligence, and scalable software that can positively impact millions of people.
Outside of work, you'll usually find me reading, playing badminton, or exploring ideas across technology, philosophy, history, and space. I'm endlessly curious about how ambitious engineering can solve meaningful human problems.
I'm always happy to connect with people building ambitious technology, advancing AI, or solving problems that genuinely matter.
Experience
Most people can't get the depth of analysis a family office takes for granted. I lead the AI and product work behind a wealth-management platform that closes that gap. It runs local-first, so a person's financial data never leaves their own machine, and every insight ships with an evidence pack that exposes the data, reasoning, and calculations behind it, so recommendations can be audited and cross-questioned rather than taken on trust. The result is analysis that's private, auditable, and rigorous, giving people institutional-grade insight they can act on with confidence.
Satellite analysis at scale was slow and manual. I built data-ingestion and deep-learning pipelines, delivered directly inside QGIS, that made the work real-time and lifted processing efficiency by 89%.
Autonomous vision fails exactly when it matters most: in fog, rain, and low light. I designed a causal-attention module that kept object recognition reliable in adverse weather, raising accuracy from a 68% baseline to 78.4%.
Open-source ML resources are often scattered and hard to learn from. Contributing to ML-Crate and Google's Bellabeat capstone, I added rigorous, well-documented analyses, reaching 88% classification accuracy and leaving cleaner references for the community to build on.
Raw data sat unused while manual reporting slowed decisions. I automated the ETL pipelines and turned their output into clear, actionable insights, cutting data-processing time by 28% and lifting customer satisfaction by 5%.
Projects
Transformer-based security scanners are too heavy to run in practice. Vuln-Eye brings CodeBERT to static Android analysis and uses Low-Rank Adaptation to cut the memory footprint by 99.2%, while holding 0.79 F1 across 60k samples.
Data stays locked behind the few people who write SQL. Ask Your Database turns plain questions into accurate SQL across MySQL and PostgreSQL, cutting manual query effort by about 90% and putting answers within reach of non-engineers.
Touch isn't always the natural way to control a screen. This maps hand movements to system volume using OpenCV and MediaPipe hand tracking, reaching 95% recognition accuracy with a real-time response of about 100ms.
Urban scenes are hard to parse from the air. Using MBRSC satellite imagery over Dubai annotated into six classes, I paired a Vision Transformer with UNet for pixel-wise segmentation, improving accuracy by 10% over prior methods at a 0.33 mIoU.
Manual attendance is slow and error-prone. Pairing YOLOv5 detection with face recognition, this marks attendance automatically at 99.38% accuracy, cutting manual tracking time by 90% across 50+ students per session.
Recommendations
“I had the pleasure of mentoring Arjit during his internship at TISeed, and I was deeply impressed by his initiative, clarity of thought, and ability to execute with minimal guidance. He created a high-quality mock application for our Entrepreneur-in-Residence program, an ambitious exercise in imagining a trillion-dollar impact venture, now published on our website. Arjit is exactly the kind of young leader we need more of: curious, committed, and capable of big thinking.”
“Arjit has clarity about his ambitions. As a mentor and advisor, I have seen the evolution of his thinking and problem-solving skills. He has a unique ability to focus on the core skills required to be an excellent data scientist, and the leadership skills to lead a team of enthusiastic people.”
“I wholeheartedly endorse Arjit for his unwavering dedication and exceptional performance during the internship. His rapid learning and proactive adaptability are truly commendable. He consistently embraced tasks with enthusiasm and a sincere commitment to excellence, an asset to any team or project.”
“Arjit has shown exceptional problem-solving skills, resolving critical issues that significantly improved our project. His code was always clean, efficient, and well-documented, making the review and merge process smooth. I wholeheartedly recommend Arjit for any team looking for a dedicated, skilled, and collaborative open-source contributor.”
Writing & Research
A book on financial clarity, co-authored with Sanjay Bhargava (PayPal founding team). It translates how the best investors think into something anyone can use.
Peer-reviewed in IJRASET. Identifies safe navigation routes across the Moon's south pole using Chandrayaan imagery.
Published by Grenze Scientific Society. Deep-learning methods to sort and classify e-waste for automated recycling.