← Rishi Sangare

My work, read one token at a time.

Each step can look back at everything before it.

reading… · arcs show how later work drew on earlier work (my judgment, not model attention)

NMIMS, Mumbai

2021 – 2026

MBA Tech: computer engineering with finance. Smart India Hackathon 2024 and 2025.

MBA Tech · CS + Finance
all case studies
CRM migrationLD Technologies · for JustRecruit (UK) · Jun – Sep 2026Moved a UK agency's 18 GB legacy CRM, with no declared keys, into a SaaS CRM that could only be written through its REST API. Nothing lost, nothing invented.253,541 activities moved with original datesLLM matching for a Japanese recruiting platformLD Technologies · for Tamago · Oct 2025 – nowRecruiters type a job description and get a ranked shortlist with reasons, in English and Japanese, fast enough to use live.0.19 → 0.57 recall@10, same model, honest golden setEvals that tell the truthAcross Tamago and RefineCV · 2025 – 2026My first result said +30%. I didn't trust a number that good, found the artifact, and reported +12.9%.+12.9% honest holdout recall@20RefineCVLD Technologies · in-house product · May – Sep 2026A B2B SaaS that turns any CV (PDF, DOCX, scans) into an agency-branded document, sold to recruitment agencies.435 commits in five monthsRecruiter CopilotLD Technologies · in-house product · 2025 – 2026A Chrome extension that scores every LinkedIn profile 0–100 against a job description, with pros and cons.11 Chrome Web Store releasesFine-tuning Mistral-7B with DPOMitwa.ai · Dec 2023 – Apr 2024With three college friends I built the backend of an AI wellbeing companion, then fine-tuned Mistral-7B when the base model wasn't good enough.7B Mistral, DPO fine-tunedThe Studio: an AI content engineMy own content business · Sep 2026 – nowOne engine runs several short-video channels: an AI agent drafts, local models render, and nothing goes out without my approval.7,400+ views on the first 6 ShortsFreelance: agent evals and automationsUpwork · Jun 2026 – nowGraded AI coding agents against real upstream bugs, and built duplicate-safe automations for small businesses.4 open-source codebases used to grade agents