Description of Current Work

​

Q: Can you describe your current position and what field/space you work in?

​​

A: My current position is a machine learning engineer, mainly in the NLP domain. NLP here is related to voice and speech—any product where speech or audio input is involved.

 

One of our active clients is MrBeast. Our product enables live translation and lip-sync to the target language. For example, if you’re watching a Korean show on Netflix, you can switch the audio to English, and our system also adjusts the on-screen lip movements to match the translated speech.

​

Originally, we built live translation for multilingual church services: attendees wore headsets and heard real-time translated audio so everyone felt included. That evolved into our current work with creators. MrBeast’s team found that audio-only dubs didn’t keep Spanish audiences engaged through the whole video. We now provide translated audio with lip-sync. (I’m not sure whether he has published with our tech yet.) Many YouTubers are clients right now.

​

Previously, I worked with ShopRite/Wakefern. During my master’s, we analyzed whether two merchandising strategies boosted sales of their private-label brands versus national brands. I did a lot of data analysis and A/B testing, and collaborated closely with the implementation/marketing teams (planograms, shelf placement, promo effects, etc.). Short project, but very insightful.

​

Typical Day at Work

​

Q: What does a typical day look like for you?

​

A: It depends on the project phase. We’re a small startup (around 10 people), so everyone touches each phase—requirements, research, modeling, integration, testing, deployment, and client conversations.

​

Early on, we evaluated combinations of translation APIs for true real-time performance (minimizing lag so lips and audio match). We later partnered with a lip-sync group and trained our own model with our own data for higher accuracy on the narrow facial region. That domain-specific training is crucial—generic models (like a general LLM) are broad, but our product needs specialized performance.

​

We deploy on AWS (after comparing AWS vs. GCP for flexibility and cost). In bigger firms, you might stay in one vertical lane; in our small team, we work end-to-end.

​

Career Journey

​

Q: Can you give a quick overview of your career journey—college, what got you interested, previous roles?

​

A: I did my bachelor’s at the Army Institute in India. I planned to join the Indian forces, but due to physical reasons it didn’t work out. My dad’s in electronics, so I always had that interest.

​

In my second year, I worked with Dassault Systèmes (aviation problems) and John Deere (agri equipment). I built image-processing + ML solutions—basic projects like object/edge detection. I wasn’t sure yet about ML vs. data science, so I took whatever opportunities came.

​

I then joined Continental (automotive, Germany). Using sensor streams (pressure, acceleration, gyro, etc.), I proposed an early crash detection approach using data before the impact—this is going into production. That convinced me to pursue a master’s. After COVID delays, I finished my master’s and moved into my current role.

​

ML vs. Data Science

​

Q: What are the differences between machine learning and data science?

​

A: Data Science is the broader umbrella: data engineering (cleaning, validating sources, removing redundancy/duplicates, fixing types), data analysis, building models, and evaluation. Think historical data, sensitivity to factors, A/B testing, probability & statistics—like predicting stock behavior using years of past data.

 

Machine Learning focuses on how models learn from data and generalize—algorithm choice, training, evaluation, deployment.

 

AI is the broader outer shell—teaching systems how to learn, not just learning a single mapping.

​

Future Career Opportunities

​

Q: What interesting career opportunities are emerging over the next 5–10 years?

​

A: Security for ML/AI (agentic AI safety): bias checks (gender, caste, location, brand), guardrails, and preventing harmful outputs—vital as AI impacts everyone.

 

Data Security/Privacy: users generate data continuously; companies must be checked/regulated.

 

Quant roles in NYC: Quant Analyst/Developer/Researcher, Statistical Analyst (strong math, stats, probability), finance/banking (fraud detection, suspicious transaction flagging). Data quality and correctness are paramount.

 

Autonomous systems: ADAS, computer vision, robotics—automotive is racing to surpass Tesla; that’s heavy AI/CV.

 

Multimodal AI: beyond LLMs/text—speech, images, sensors combined. Humans are multimodal; systems need to be, too.

​

Advice for Students

​

Q: Any specific advice for someone interested in your field?

​

A: Start early and focus on mathematics—especially probability and statistics. Clear basics let you see problems differently and choose the right approach. Learn some Python and SQL. You can’t avoid some programming in this field, but we’re not doing software-only work all day. Keep learning continuously—quarterly self-checks on your skills help prevent becoming irrelevant.

​

Calculus and Advanced Math

​

Q: Do you use advanced math like calculus? When would you need it?

​

A: Yes—calculus shows up (e.g., area under curves that a model depends on), though often as a small but critical part of an algorithm. You should understand what is being computed and why. Emphasize probability, statistics, and then add calculus/integration/differentiation as the problem demands.

​

Cutting-Edge Aspects

​

Q: What about your work feels cutting edge?

​

A: Everyone is focused on LLMs, but the real frontier is multimodal AI—text + audio + images + sensor data, together. That’s closer to how humans perceive. LLMs are important, but things will shift; you don’t want three years of only LLM work and then feel redundant. Our real-time translation + lip-sync is a concrete multimodal challenge at the edge of NLP, speech, and vision.

​

Goals

​

Q: What are your short- and long-term goals?

​

A: I don’t want to lock into a single industry; the same domain skills can apply across many. I want to see how the data/ML domain plays out in different industries and keep learning continuously. Personally, I track my growth (e.g., quarterly skill audits) to avoid stagnation.

​

Favorite Part of the Work

​

Q: What’s your favorite part about your work?

​

A: Connecting with people and staying active in the community. NYC has tons of AI/ML meetups, product demos, and talks. Join LinkedIn groups, attend local events, try early releases, hear first-hand about roadblocks and roadmaps—that keeps you current and plugged in.