vyntra AI LABS · CAREERS — 2025
Build the next
instrument.
Small teams, physical-world problems, and code that runs where the trucks are. Your model ships or it doesn't — you'll know by Friday. The map on the right is the team: five disciplines, 120+ engineers, and one dashed node that's still open.
FIG. 00 — LIVE TEAM MAP · 5 DISCIPLINES ORBIT THE PLATFORM CORE · PACKETS ARE REAL EVENTS · THE DASHED NODE IS THE OPEN SLOT. HOVER A CLUSTER — CLICK TO SEE ITS ROLE.
01 — OPEN ROLES
Open right now.
Four ways in. Each one puts real hardware, real data and a real pager in your hands — the interview is honest about that, and so is the job.
4 ROLES LIVE23 APPLICATIONS THIS WEEKMEDIAN FIRST REPLY 31 HNEXT BATCH START JUL 07
You'll turn 40M+ events a year into decisions: which sites leak dwell time, which gates underperform, what next quarter looks like. Full-time on the Data team — the metrics layer you build becomes the one the whole company reads.
YOU WILL
- Own dashboards that plant leaders actually open every morning
- Run deep-dives on site performance and model outcomes
- Build the metrics layer (SQL + dbt) the whole company reads
- Present findings to customer ops teams monthly, in their language
YOU BRING
- SQL you're proud of, Python (pandas) you can defend
- Comfort with Metabase / Superset / Grafana
- 1–3 years in analytics or data engineering
- Bonus: logistics, manufacturing or IoT sensor data
Our gates and bridges run 24×7 — so do you, in rotation. You'll debug vision pipelines remotely, own uptime SLAs across your time zone, and be the calm voice when a site in Dammam is down and a truck is waiting. Full-time on the Reliability team.
YOU WILL
- Triage and resolve site incidents — cameras, edge boxes, models
- Own uptime SLAs across your time zone
- Feed patterns back to engineering: bug → fix, not bug → ticket closed
- Write runbooks until your shift is boring
YOU BRING
- Linux confidence and basic Python
- Docker and networking fundamentals
- Calm written English under a ticking SLA
- Willingness for a paid on-call rotation
Train, break and shrink models that read the physical world — then watch them run on Jetsons at factory gates, in monsoon light, at 40 km/h. Full-time on the Vision core team. Week one is reading code; week two is shipping to a real gate.
YOU WILL
- Train and evaluate detection / ANPR models on real, messy footage
- Optimise for the edge — TensorRT, quantisation, INT8
- Own eval datasets and regression suites
- Ship to production weekly; carry the pager for what you ship
YOU BRING
- PyTorch and OpenCV fundamentals
- One shipped project — link it, that's your interview
- Comfort with imperfect data and honest metrics
- 0–2 years; degree optional, evidence isn't
Six months inside a real vision stack: data tooling, model evaluation, edge benchmarks on actual Jetson hardware. Most interns convert to full-time; all of them leave with a portfolio that isn't a TODO app. Stipend ₹25,000/month, batch starts July.
YOU WILL
- Build eval and labelling tooling used by the whole team
- Benchmark models on Jetson hardware — real numbers, real boards
- Shadow a deployment end-to-end, gate to ERP
- Present your own mini-project at month five
YOU BRING
- Final-year student or recent graduate
- Python plus genuine curiosity
- Portfolio over GPA — show us something you built
- Availability for six months, on-site in Bengaluru
02 — WHERE YOU FIT
Start with what you love.
Titles are shorthand. Find the sentence that sounds like you, and click it — we'll take you to the role.
03 — HOW WE HIRE
Four steps, no mystery.
You will always know which step you're on and what happens next. We've been the candidate on the other side of a black-box process — we swore.
Apply
One short form — no cover-letter theatre. Link what you've built; that's the real CV. Acknowledged within 48 hours, always.
Conversation
Thirty minutes with the team lead. It runs both directions — bring your hardest questions about the stack, the pager and the pay.
Take-home
Time-boxed to four hours, your tools, a real slice of our data. An engineer reads every submission line by line — no keyword filters.
Team day & offer
Half a day with the crew — pair on something real, meet the team, see the lab. Offer within 48 hours of the team day.
04 — LIFE AT vyntra
Serious work, human pace.
We're small enough that your first PR ships in week one, and stubborn enough about quality that it gets reviewed properly anyway. Teams own their models in production — including the pager.
And yes: you will stand at an actual factory gate at some point, watching your code wave a real barrier arm for a real truck. Nobody here has gotten tired of that yet.
Before you sign, you spend a paid shadow day with the team — the offer is easier to judge from the inside.
05 — FROM THE FLOOR
What it's actually like.
“I shipped a model to a real gate in week two. My friends from college are writing TODO apps.”PRIYA S. · ML ENGINEER, BLR — 2024 HIRE
“The on-call pager is honest: it pays, it rotates, and it respects sleep. That's rarer than it should be.”ARJUN M. · RELIABILITY, BLR
“Nobody asked where I studied. They asked what I'd broken, and what I'd learned from it.”MEERA K. · DATA, BLR — INTERN → FULL-TIME
06 — ASKED OF US
Asked, answered.
No. Two of our strongest engineers came from physics, one from a bootcamp, and our best data analyst started in operations. We read what you've built, not where you studied. Evidence over pedigree, every time.
AI Tech Support runs one seat remote for the US time zone. Everything else is Bengaluru — on-site or hybrid — because the work is physical: cameras, barriers, weighbridges and the occasional dusty site visit. You can't debug a gate you've never stood at.
Four hours, honestly time-boxed — we'd rather see your ceiling in four hours than your weekend. It's a real slice of our data, in your tools. An engineer reads every submission and the code review you get back is real, even if the answer is no.
Median offer-to-first-commit is eighteen days. Interns batch in January and July; full-timers start when they're ready — we hold offers up to eight weeks for the right person.
No role that fits?
Write to us anyway — we keep good people on file, and roles open faster than we post them. Six of this year's hires applied for a job that didn't exist yet.