Applied AI Engineer

Why we exist

People will always sell, advise and serve. Someone deciding on a loan, an insurance policy or an investment wants a person across the table. That is not going to change.

What can change is how ready that person is. Today, how good a field agent becomes depends on how good their manager is. The best managers play the difficult customer, coach every conversation and know exactly who is ready. There are never enough of them.

UpTroop gives every frontline person the coaching of the best manager on the team. AI does not replace people here. It helps them move up the value chain.

Where we are today

• 14,000+ frontline people practise on UpTroop in production, at financial services companies in India.
• Voice-first AI role-plays and coaching in 11 Indian languages, delivered on WhatsApp, Microsoft Teams and the web.
• Every practice conversation is scored and coached, and leaders see readiness by agent, branch and region.
• We are now taking UpTroop to insurance distribution in the US.

Where we are going

We are not a thin wrapper around someone else's model. The hard problems are ours to solve:

• AI avatars that feel like real customers. Lifelike personas you can interrupt, argue with and win over: the business owner who has heard five insurance pitches this week, the first-time borrower who is nervous about the paperwork.
• Coaching beyond words. Today we coach on what was said and how it was said. Next comes pace, tone and confidence, and eventually body language and presence on video.
• Speech that works for India. Low-latency voice across Indian languages, regional accents and people switching languages mid-sentence, where off-the-shelf models still struggle.
• AI at the edge. Field agents use mid-range Android phones on patchy networks. Moving parts of the voice and scoring stack onto the device is a problem we want to crack.
• Evaluation as a product. Scoring a sales conversation the way a great manager would, consistently, across thousands of people and many languages.

About the role

You will own the intelligence behind UpTroop: the AI customer an agent practises with, the coach that gives feedback, and the scoring that decides who is ready. This is production work used every day by thousands of people, not a research project or a demo.

What you will build

Real-time voice conversations

• Speech-to-text, LLM and text-to-speech pipelines with low enough latency to feel like a real call.
• Natural turn-taking, interruptions and back-channelling across Indian languages and accents.

AI customers and avatars

• Personas that stay in character, push back with realistic objections and react to what the agent actually said.
• The move from voice to visual avatars, and from words to tone and presence.

Scoring and coaching

• Rubric-based evaluation of sales and compliance conversations that a sales head would trust.
• Evaluation pipelines, test sets and regression checks so every model or prompt change is measured.

Knowledge to practice

• Pipelines that turn product notes, SOPs and disclosure rules into role-plays and training content.
• Retrieval and grounding so the AI never invents a product feature or a policy rule.

Cost, scale and the edge

• Model routing, caching and observability across many customers and languages.
• Experiments with smaller and on-device models for low-bandwidth field use.

What we are looking for

• 4-5 years of software engineering, with at least 2 years shipping ML or LLM systems to production.
• A strong pedigree: IIT, NIT, BITS, IIIT or similar, or a track record that speaks for itself.
• Excellent Python and solid computer science fundamentals.
• Hands-on experience with LLM APIs and open-source models, prompting, tool use and evaluation.
• You care about latency, reliability and cost, not just whether the demo works.
• You want to grow with an early-stage company and take ownership of a whole area.

Bonus points

• Speech recognition, speech synthesis or real-time audio (WebRTC, streaming).
• Multilingual or Indic language models.
• Computer vision, avatars or multimodal models.
• Model optimisation, quantisation or on-device inference.
• Open-source contributions or published work.

What we offer

• Compensation: ₹15 LPA fixed plus meaningful ESOPs. We are looking for people who want to build long-term ownership in the company, not just a salary.
• Location: Bengaluru, hybrid. Two days a week in the office.
• Founders, not layers: you work directly with the founders.
• Early-team ownership: you will shape how we build, sell and scale, and grow into a leadership role as the company grows.

How to apply

Email your CV to hello@uptroop.io with the subject line Applied AI Engineer. Add a short note on the hardest AI system you have shipped and what you would build first at UpTroop.