Now
- Working on my startup Convertive by Outhad AI: Re-defining the way your customers interact with your store. 🎉
- Building Franz: a functional programming language with Rust-like safety and a tiny, keyword-free functional core — LLVM native compilation, lexical scoping, and deterministic replay.
- Exploring Transformers, Diffusion Models, LLMs, Computer Vision, Machine learning, and Android/IOS at Beloit.
- Working on some fun side projects and busy writing something... 🎉
About
I'm primarily a technical builder and serial founder — built 2 startups Outfly and FractalXR (acquired by Boxally). My academic background is in applied Math, CS and Quantitative Economics, with splashes of philosophy, physics, and psychology.
Lately I've been designing Franz, a functional programming language built in C with LLVM native compilation — a tiny, keyword-free functional core with prototype-oriented objects, Rust-like safety, lexical scoping, and deterministic replay.
I spend a lot of time building Convertive by Outhad AI, reading works by Jeffrey Archer, Jaun Elia, and J.K. Rowling, and playing and watching cricket.
Experience
- genera8tor(gBeta) incubated startup. Working on improving the way you search for retailer to end users, bridging the gap between what you speak to what you get by leveraging generative AI, computer vision, natural language processing, and machine, and deep learning.
- Incubated by genera8tor.
- Built scalable ads auction system in Go for handling 10M+ daily requests at p95 under 120ms.
- Maintained 99.9% auction availability across multi-AZ Kubernetes deployment and failover.
- Shipped feature-flag service powering checkout experiments with automated rollout controls.
- Reduced PostgreSQL load 40% with Redis caching for sub-millisecond metadata lookups.
- Processed billions of events with Airflow pipelines, distributed tracing, and automated recovery.
- Selected from 400+ applicants for a competitive 10-week program focused on enhancing SE skills through hands-on projects, interview prep, and technical workshops.
- Received 1-on-1 mentorship from Google SWEs; collaborated with peers on innovative solutions and gained exposure to industry best practices.
- Innovated noise-aware 3D U-Net model, improving PET image quality by 15%, surpassing U-Net performance.
- Worked on AG-PET: Anatomically Guided Multimodal U-Net for PET denoising.
- Constructed patch-based NAFNET: Noise-Aware Fusion Network.
- Worked on Self-Supervised Diffusion model to make the diffusion model noise aware to improve the accuracy of denoising.
- Presented research findings at the IEEE NSS Medical Imaging Conference 2024 in Tampa, Florida.
Supervisor: Ezzat Elmoujarkach, Magdalena Rafecas
- Provided custom AI solutions to companies from AI marketing to branding, and data-driven solutions for businesses of all sizes.
- Primarily focused on the Financial and Ecommerce sectors of businesses.
- Specialized in helping small to medium-sized enterprises grow efficiently and cost-effectively.
- AI-powered strategies elevate customer engagement and revenue, simplifying operations along the way.
- Developed a Python-based desktop application for automating tasks and saved $10K+ by in-house production of custom Financial Aid letters for each student, previously done by outsourced services.
- Developed a Python-based mobile system for 3D face reconstruction from 2D images, aimed at supporting cancer and cosmetic surgery planning.
- Leveraged a GAN trained with PyTorch, achieving 90.2% accuracy rate on the LFW Dataset.
- Utilized DLib to identify 68 key facial landmarks and employed SciPy, PyMesh, and Matplotlib.
Supervisor: Dr. Ganesh Mani
- Developed and launched a server using Google Cloud OAuth 2.0 on AWS ECS for ML lifecycle management.
- Authored Python scripts for monitoring Mutual Fund Scheme Expense Ratios.
- Engineered AWS Lambda Functions for verification of investor information across 20+ Asset Management firms.
Research
Representation, Benchmarks, and an Honest Off-Policy Study. Preprint, 2026 · Zenodo · DOI 10.5281/zenodo.20652076.
Propensity-logged benchmarks for e-commerce intervention bandits. Dataset, 2026 · Zenodo · DOI 10.5281/zenodo.20651686.
Updates
One of my more recent goals is to send out personal life updates every few months to those who are interested. If you're interested, fill out this form.
I also occasionaly write about topics I find interesting at Medium.
Projects
Schema contract verification for dbt — catch breaking changes before they break production. A static analysis tool that infers schemas from SQL and compares them against declared contracts to flag breaking changes before deployment.
A high-level, functional, interpreted programming language with Rust-like safety and a tiny, keyword-free functional core with prototype-oriented objects. Features LLVM native compilation, lexical scoping, a standard library, and deterministic replay.
AI-powered options trading platform built as an industry-standard quant system. Integrates 37+ ML models, advanced Greeks calculations, volatility surface modeling, reinforcement learning for dynamic hedging, and backtesting with real-time risk management.
Professional-grade telemetry analysis platform for Formula 1 games. Streams live UDP telemetry into 12 interactive dashboards with AI-powered predictions for tire wear, lap times, pit strategy, and race outcomes, persisting data to SQLite.
AI agent observability and safety platform with formally verified mesa-optimization detection. Provides tracing and monitoring for autonomous agents while detecting inner alignment failures through mathematical proof certificates.
Ultra-low latency tracing system for performance-critical applications. Combines Linux eBPF/XDP and io_uring for kernel-level tracing with sub-microsecond overhead, targeting high-frequency trading and latency-sensitive workloads.
Deployable Cloudflare Worker that sits in front of LLM provider traffic, applying deterministic normalization, exact cache, guarded semantic retrieval, rollout controls, and provider-aware safety checks before forwarding misses upstream to OpenAI, Anthropic, and Gemini.
The system of record for regulated AI agent actions. An SDK and collector that records agent actions into an append-only, hash-chained store — replayable, queryable, and tamper-evident — with policy enforcement and evidence export, no stack replacement required.
Built ultra-high performance HFT network analyzer with custom memory pool achieving 2.27M ops/sec and system throughput of 127K+ messages/second. Features real-time arbitrage detection across NYSE/NASDAQ/CBOE, sub-microsecond latency optimization, lock-free queues, and regulatory compliance for production trading environments.
Engineered production-grade FX arbitrage engine achieving sub-20s latency. Delivered 99.5% performance optimization, 1M+ msg/sec throughput, and enterprise risk management across 16 institutional liquidity providers.
Built a backend cursor for 3D modelling with a custom physics engine. A game blending Clash of Clans + Minecraft + Cursor for 3D modelling with building physics.
A High-performance Compiler Accelerator for LLM and Linear Algebra Operations.
Semantic data processing Python package integrating LLMs with DataFrame operations for intelligent data analysis. Implements modular semantic operators (search, transform, cluster) with multi-modal processing support and automated index management.
Python framework for multimodal analysis, specializing in synchronized video-audio processing, emotion detection, and LLM-enhanced analysis.
A powerful middleware package that provides a unified interface for executing SQL, Cypher (Neo4j), and vector similarity queries.
Create and deploy Voice AI Agents to qualify leads, boost sales, automate customer support, and streamline recruitment and hiring.
Cursor for storytelling — instantly turn your sketches into animated 3D worlds with AI.
A Rails engine for recording, replaying, and debugging HTTP requests with their database interactions in real-time.
Powerful TypeScript/Babel compiler plugin that enforces accessibility best practices at compile time, catching common accessibility issues early and ensuring WCAG compliance.
Transform any document into a dynamic template and generate customized versions in seconds, saving you hours of manual work.
A comprehensive quantitative analytics platform providing insights into federal contract data and spending patterns for investment analysis and risk management.
Trained high-accuracy hand gesture classifiers by transfer learning from VGG-16 and ResNet-50 CNN architectures.
Fashion outfit generator powered by GenAI and LLMs, leveraging user order history and cart data to provide personalized outfit recommendations.
University of Trading Competition 2023. Financial analysis tools for optimizing portfolio weights using genetic algorithms, evaluating risk with CVaR, and maximizing Sharpe ratio for best risk-adjusted returns.
Multi-modal classification system that leverages both text and image data to identify complaints in reviews.
AI-powered OS browser to search and navigate the web and manage file explorer with a built-in terminal.
Real-Time Exercise Monitoring System with Pose Detection using Mediapipe and OpenCV.
End-to-end poetry generation from a list of topics using T5: Text-To-Text Transfer Transformer.
Agentic AI-based Code Editor.