VR campus tour
Campus 2050
Stall concept and pitch for a VR tour of the college in 2050. Boarding-gate set, image direction, and judge narrative.
Portfolio / 2026Mumbai, IN · 19.0760° N, 72.8777° E
Swayam Siddharth WaghchaureProduct engineer/Mumbai, India
I build for themoment thingsgo wrong.
Interfaces and systems for the last mile — the empty state, the failed upload, the check-in that never arrives.
Selected work / 2026
Three projects
One pattern: start at the failure state, then design the shortest path back to confidence.
Chapter 01 / 03Team project · 2026
TV-first recommendation app · Amazon Vega / Fire TV
YourNextBinge
Interface study960 × 540
D-pad onlyYour Next BingeMood / 01
Pick a mood.
83 titles · 7 moods · offline
Recreated from the Vega buildClick a screen to move
The problem
A TV remote is a terrible search tool. Endless grids, no keyboard, and twenty minutes of scrolling turn “what should I watch?” into its own evening.
What I built
A 960×540, D-pad-only app that turns 83 titles and 7 moods into a five-item shortlist, with a one-line reason for every pick. The catalog, scoring, and explanations all run locally.
Decisions
- 01Deterministic recommender: weighted scoring, no ML, no randomness, no network.
- 02Fixed relaxation ladder — intensity, then genre, then era, then runtime. The UI never shows an empty list.
- 03A predictable first focus on every screen, with focus maps wired explicitly for row edges.
- 04Jest coverage for scoring, relaxation order, navigation, and rendering.
Result
A complete TV flow that answers the question in one screen and can explain every recommendation it makes.
83
titles in the catalog
7
moods
4
screens, D-pad only
0
network calls
Chapter 02 / 03Team of 4 · 2026
Fault-vs-noise classifier · SMLRA × Abhiyantriki 2026
Rogue Protocol
Validation setn = 1,000
Threshold 0.54One cell = one log · arrangement illustrative
True fault, caught 138Real fault, missed 71False alarm 57Noise, ignored 734
Model comparison
LogReg floor0.6022
LGBM challenger0.6462
HGB · promoted0.6832
+0.081 over the LogReg floor. Errors only dims the 734 true negatives.
HGB + SVD-64 + keyword flagsTeam result · 4 people
The problem
A ship computer floods the crew with false alarms and hides real faults inside calm log lines. Every wrong verdict either wakes the crew for nothing or lets a real failure pass.
What I built
A classifier trained on 6,000 labelled logs, tuned to catch real faults without crying wolf. Validation F1 landed at 0.6832 at threshold 0.54 — 0.081 above the LogReg floor.
Decisions
- 01HGB + SVD-64 + keyword flags; the LGBM challenger was rejected at 0.6462.
- 02Threshold sign-off and a checker-gated CSV. Every upload had to print TRANSMISSION VALID.
- 03Round 2 orchestration across triage, memory, scheduling, and notifications.
- 04Handover and timeline reports generated from real traces, not hand-written summaries.
Result
3,000 unlabelled logs classified, with submission integrity held from threshold to upload. Team result.
0.6832
validation F1
+0.081
over the LogReg floor
3,000
unlabelled logs classified
4
person team
Chapter 03 / 03Solo · BUILX · 2026
Night-travel safety app · Student commuters
Quiet Check-in
BUILX / Solo buildLocal-first
StudentApp
Quiet Check-in
Night travel
Route50 min · ₹640
6.2
Take care
Cab via Eastern Express Hwy
Weakest stretch: JVLR flyover · 5.4/10
CheckpointsDue 00:10
Hostel (Powai)23:30
JVLR flyover23:40
EEH (Ghatkopar)23:50
Sion Circle00:02
CSMT00:20
Weakest area / factorsJVLR flyover
Lighting7
Footfall3
Help nearby5
Main road10
Open shops1
Contact's phoneWhat they see
Contact view23:42
On the way home
Swayam is
moving.
Cab via Eastern Express Hwy. No live map, no constant pings — just the next check-in.
Last check-in23:42
Next check-in00:10
Arrival00:20
If the 00:10 check-in is missed, you'll get the route and last known location at 00:20.
Silent by default
No notification while the trip is on track. One alert only when it matters.
Route score = 70% average + 30% weakest stretchScores are estimates, not official data
The problem
Getting home at night is a chain of small uncertainties. Sharing live location all the time is noisy; doing nothing is worse.
What I built
A local-first app that stays silent while a trip is on track and alerts a contact only when a check-in is missed. Built solo for BUILX, the college tech club event.
Decisions
- 01Five steps: Setup, Routes, Plan, Live, How. No account, no backend.
- 02Day and night safety scores across lighting, footfall, main-road access, emergency access, and open shops.
- 03A route is only as safe as its weakest stretch: the score weights the minimum area above the average.
- 04The contact view shows only what matters — route, last check-in, next deadline — not a map full of noise.
Result
A complete solo prototype with a full trip flow, a working contact view, and no server to depend on.
5
safety factors
6
transport modes
5
step flow
0
backend calls
About / Method
Mumbai, India
I start with the failure state.
Engineering undergraduate at MIT Mumbai Engineering College, expected Nov 2030. I work where interface design meets applied ML.
Most of my attention goes to the last mile — the empty state, the failed upload, the one sentence that tells someone what to do next. Then I design the shortest path back to confidence.
Toolkit
What I reach for
Interface
- React 19
- React Native
- Next.js 16
- TypeScript
- Tailwind CSS 4
TV
- Amazon Kepler 4.0
- Vega OS
- D-pad focus
- 960×540 layout
Data
- Python
- pandas
- NumPy
- scikit-learn
- HGB
- LogReg
- SVD
- TF-IDF
Practice
- Git / GitHub
- Jest + Testing Library
- ESLint
- Linux / Fedora
- Technical writing
Education
MIT Mumbai Engineering College
Degree
Engineering undergraduate · Computer / IT
Graduating
Expected Nov 2030
Also / Explorations
Not counted in selected work
Cloud kit
Pi Home Server
Plug-and-play Raspberry Pi cloud kit. Campus pre-order flow, concept document, and an Apps Script form and sheet.
Contact / 2026
Mumbai, India
19.0760° N, 72.8777° E
Open to internships
The next step
Have a problem that has to hold up when things go wrong?
Mumbai / 19.0760° N, 72.8777° E
Open to internships