SOFTWARE ENGINEER IN TRAINING.
hi, i'm sachin. cs undergrad at the university of leeds (meng, class of 2027), aspiring software engineer. i build backend systems and applied ai tools, with a strong focus on testing and reliability. last summer i shipped ai workflow automation inside kpmg as an intern; now i'm ai/ml project lead at the university's ai society.
REPOLENS READS YOUR REPO. ANSWERS WITH RECEIPTS.
a deployed ai code-intelligence platform with 27 pre-indexed repositories and 60k real source chunks. classic rag and a tool-using google adk agent answer codebase questions with file-and-line citations, backed by fastapi, postgresql/pgvector, gemini, neon, docker, and 30 automated tests.
[+] retrieved 5 chunks · 2 files cited
[+] answer grounded in README.md:1-14
▋
FORKLAB RUNS FIVE FIXES AT ONCE. YOU KEEP THE BEST.
a hackathon-winning parallel ai sandbox platform. it orchestrates multiple browserpod execution environments to generate, test, and compare software solutions safely — llms plan the changes, sandboxes execute them, and humans review proof before anything is approved. built workflows for a/b frontend variant testing and fixing github issues in parallel.
THREE TIERS. REAL-TIME BIDS. 95% TESTED.
an online auction platform designed as a three-tier system with role-based access control and real-time bidding. i built rest endpoints and the relational schema, implemented ci/cd with github actions and docker, and drove test coverage to 95% with unit and integration tests — inside a scrum team of five running standups, sprint planning, and code reviews over eight weeks.
[ci] docker build ✓ · 41 unit + 17 integration ✓
[ci] coverage: 95% · deploy: green
▋
HOW WELL DOES SAM2 SEE THROUGH BAD WEATHER?
a robustness evaluation of segment anything model 2 under synthetic adverse-weather corruptions in urban driving scenes. i built a controlled pipeline over a cityscapes subset, applied five corruptions — rain, fog, frost, snow and brightness — at increasing severity, and measured how far sam2's masks drifted from its own clean-image output.
consistency fell as severity rose: frost degraded it hardest, then heavy snow; fog and brightness were milder, while rain behaved non-monotonically and scene-dependently.
SHIPPED AI TOOLING INSIDE A BIG FOUR FIRM. AS AN INTERN.
at kpmg i built and deployed a full-stack ai application — fastapi, next.js, postgresql on neondb, openai-based workflows — automating developer processes across github and jira. i designed the backend apis and multi-step orchestration logic for structured execution and validation, shipped features in fast iterations across vercel, railway, and neondb, and hardened reliability through testing, debugging, and release fixes.
RUNNING THE AI SOCIETY'S BUILD PROGRAMME.
elected ai/ml project lead at the leeds ai society (aisoc), aug 2026 – now.
the 2nd annual leeds ai week ran five days of panels and workshops on business, education, tech, engineering and the environment, with speakers from industry and academia.
aisoc's first hackathon, run overnight in collaboration with the openai student collective.
judged by academics and industry on creativity, technical depth, impact, design and responsible use of data. prizes were sponsored by the school of computer science, cdt and lida.
CERTIFICATIONS.
the paper trail — training completed, programmes finished, and awards won along the way.
BOULDERING
1.5 years in, currently climbing V5–V6. competed in the IFSD Bouldering Cup. same loop as debugging — read the problem, fail it, stick the move on the next go.
GUITAR
6–7 years playing, distinction in Trinity College London Grade 1 music theory. the thing i do to slow my brain down after a long build.
GAMING
competitive shooters (valorant, marvel rivals, fortnite, brawlhalla) and casual runs (roblox, meccha chameleon) — plus story-mode favourites: 007 first light, uncharted 4, shadow of the tomb raider.
WEEKENDS SPENT SHIPPING.
the fastest way i learn is a deadline and a team. every entry here went from empty repo to a working demo inside a weekend.
-
NOV 2026
KEVIN
NATWEST TRACKUNIVERSITY OF MANCHESTER · GREATUNIHACK · TEAM OF 4
a voice that talks students on loans out of impulse buys. kevin is a chrome extension that peeks in on amazon and asks, out loud, why you need it. an elevenlabs voice agent checks your answer against the budget you set, while firecrawl hunts for the same model cheaper at a uk retailer. it never blocks — you always decide.
-
MAY 2026
FORKLAB
★ WINNERUNIVERSITY OF LEEDS · “AI IN THE BOX”
don't trust ai-written code — prove it. forklab boots a disposable browser sandbox, writes the patch into it, runs the tests inside, and streams the result back. the demo takes a failing csv-export test, applies a patch, reruns it green, and emits a proof report before a human approves.
-
MAY 2026
DELT
VERCEL · ZERO TO AGENT · TRACK 2 (v0 + MCPs)
docs rot the moment code moves. delt watches pull requests, works out which docs the change invalidates, pulls extra context through the context7 mcp, and drafts a reviewable patch — then waits for a human before opening the issue. every code delta, documented.
-
OCT 2025
AI HACK LEEDS
< 6 HOURSLEEDS DIGITAL EDUCATION × PLATO AI · TEAM OF 2
an ai study platform that turns a static lecture pdf into something you can actually revise with: auto-generated multiple-choice quizzes, a conversational study copilot grounded in your notes, and 20-second explainer video storyboards. built entirely on gemini — no paid extras — in under six hours.
-
MAR 2025
GREEN LINK
QUACKATHON · UNIVERSITY OF DUNDEE · 24H
two challenge tracks in one build: a web platform that connects like-minded people around volunteering, with a unity game embedded straight into the browser as the reward layer. shipped an ai chatbot and got webgl talking to react in a day — the embedding was the fight, and the win.