hi, i'm lakshit sachdeva
i build machine learning products, retrieval systems, and weirdly specific tools that feel useful the second you touch them. currently spending time at ey and shipping things on my own.
most of what i like sits somewhere between software, ml, and product taste. private memory, numeric reasoning, agent workflows, interfaces that make complicated systems feel obvious.
also shipping experiments at athair and recurdo.
now
building at ey
base
mumbai
focus
retrieval, reasoning, local-first systems
side quests
research, products, open source
selected work
Recurdo - local-first AI memory engine
A private memory layer that indexes files, screenshots, browser activity, chats, and saved links so information can be retrieved through contextual natural-language search instead of exact filenames.
Athair - trade thesis stress testing
An AI-powered engine that pressure-tests trade ideas with thesis decomposition, volatility regime detection, fat-tailed Monte Carlo simulation, and confidence-driven invalidation logic.
Smart waste management system
A hardware-software co-design build with a fine-tuned ResNet-50, OpenCV-based live inference, Streamlit monitoring, and Arduino-driven servo actuation for automated waste segregation.
Financial portfolio management system
A full-stack tracker with secure JWT auth, multi-asset portfolio views, real-time performance analytics, and Supabase-backed live updates for holdings and allocations.
CAPTCHA-X benchmark study
A synthetic benchmark reliability study showing CAPTCHA recognition performance can collapse under cross-generator transfer, backed by a 160,000-image benchmark and reliability-aware evaluation.
toolkit
ml / ai
backend / data
frontend / product