Production ML & LLM systems
Go beyond notebooks—learn how retrieval, agents, and model APIs behave under load, cost limits, and real user traffic.
RAG pipelinesAgent toolchainsOffline & online evalsPrompt versioning
We're hiring·2 open roles
Grow as an AI engineer on work that ships
Vegrade is a principal-led team building production data science and AI systems for startups and growth companies. Interns and full-time engineers learn by delivering — mentorship, client-grade problems, and the discipline of evals, observability, and handoff.
In the form or email, note Internship or Full-time — we route careers separately from project inquiries.
6–12
Week typical client programs
Fast cycles, real shipping
1:1
Mentor ratio on intern cohorts
Staff engineer pairing
100%
Production-minded delivery
Evals, observability, handoff
Outcomes
What you'll learn and carry forward
Whether you join for a summer or a career, the goal is durable skill — not resume keywords. These are the capabilities people build here.
Data science in the real world
Frame problems with metrics, build reproducible experiments, and understand when classical ML beats generative approaches.
Experiment designFeature pipelinesModel promotionDrift awareness
Engineering craft
Write code your teammates can maintain: typed APIs, tests where they matter, CI gates, and architecture notes that survive handoff.
TypeScript / PythonAPI designObservabilityCode review discipline
Client-ready communication
Practice the skills consultants and staff engineers use daily—scoped updates, risk flags, and demos that tie to business outcomes.
Written RFCsWeekly readoutsScope & change controlRunbooks
Programs
Internships and full-time tracks
Both paths emphasize mentorship, written communication, and production discipline. Headcount stays limited so every person gets meaningful scope.
10–14 weeks
Internships
Applied AI engineering
A bounded project with a named mentor—design through demo week on a stack used on real client work, not a toy repo in isolation.
Typical outcomes
- Ship a scoped feature or eval harness in a production-style codebase
- Present tradeoffs in a final readout (architecture, metrics, risks)
- Receive structured feedback on code, communication, and ownership
- Optional return offer for strong performers when headcount allows
Good fit if: Students and early-career builders with solid CS fundamentals and curiosity about applied ML/AI.
Ongoing
Full-time
Mid-level and senior
Own workstreams across data, models, and product surfaces—partnering with principals and client engineering leaders with minimal hierarchy.
Typical outcomes
- Lead delivery on agents, RAG, MLOps, or full-stack AI product lanes
- Shape technical direction within fixed-scope programs
- Mentor interns and contribute to reusable playbooks
- Grow toward staff-level scope through judgment, not tenure alone
Good fit if: Engineers who have shipped something real and can explain why they made hard technical choices.
Roles
Where we're hiring
Two open internships — Vachai AI research and growth — plus delivery roles coming later this year. If your profile is close, reach out even if the title doesn't match exactly.
- Open nowInternship· Vachai AI · Research
NLP & Data Intern — Vachai AI
Project: Vachai AI — dialect-aware Telugu foundational model
Join the team building Vachai AI—a dialect-aware Telugu foundational small language model (SLM). You will help source diverse Telugu speech and text, engineer pipelines that turn raw data into training-ready corpora, and learn how those datasets feed pre-training and fine-tuning.
Telugu NLPData pipelinesSLM trainingFine-tuningWhat you'll work on
- Collect and curate Telugu audio across regional accents and speaking styles, with clear metadata (dialect, region, speaker context where appropriate).
- Gather English-keyboard (Romanized) Telugu conversation data—chat-style text that reflects how people type Telugu online.
- Design and implement data pipelines: ingestion, validation, deduplication, and versioning for speech and text sources.
- Apply NLP preprocessing—normalization, transliteration handling, tokenization choices, quality filters—and document decisions.
- Deliver cleaned, auditable datasets ready for SLM pre-training and evaluation splits.
What you'll learn
- End-to-end data engineering for language models—not just labeling, but reproducible pipelines.
- NLP techniques for low-resource and dialect-rich languages (Telugu variants, code-mixed text).
- How pre-training data quality affects model behavior—and how to measure it.
- Hands-on exposure to training and fine-tuning workflows for small language models (with mentor oversight).
- Experiment tracking, eval basics, and responsible data practices for speech and text.
- Open nowInternship· Growth · Open now
Marketing & Sales Internship
Join the growth team to promote Vegrade's AI engineering services and generate qualified leads. You will work directly with the founders on outbound experiments, content for technical buyers, and the early sales motion—learning how a principal-led services company builds a pipeline from zero.
OutboundContentLead generationSales opsWhat you'll work on
- Identify and research target accounts (founder-led startups, AI-adjacent product teams) and build prospect lists with clean firmographic data.
- Draft outbound sequences across email and LinkedIn—copy tuned to engineering and founder audiences, not generic templates.
- Help produce short-form content (case briefs, technical explainers, social posts) that promotes Vegrade services and reference programs.
- Track the pipeline in a CRM: response rates, qualified conversations, conversion to discovery calls, and reasons for drop-off.
- Coordinate with engineering principals on positioning, messaging experiments, and feedback loops from real sales calls.
What you'll learn
- How a small AI engineering firm acquires its first customers—what actually works, what burns money, and how to measure both.
- B2B copywriting for technical buyers: engineers, CTOs, heads of product, and founders.
- Outbound tooling and CRM discipline (Apollo, HubSpot, or equivalent) and how to instrument a sales funnel from scratch.
- Enough fluency in services like RAG, agents, evals, and MLOps to talk to prospects clearly—without overclaiming.
- Prioritization and reporting frameworks used by growth and sales teams at modern services companies.
- ClosedFull-time· Closed
ML / AI Engineer
Build and harden models, retrieval stacks, and eval harnesses on client programs.
PythonRAG or agentsMLOps basicsProduction debugging - ClosedFull-time· Closed
Full-Stack AI Engineer
Ship APIs, web surfaces, and AI features with clear boundaries between product and model layers.
TypeScriptNext.js / FastAPILLM integrationObservability - Coming soonFull-time· Coming Soon
Data Engineering (ML)
Own pipelines, curated tables, and feature readiness for modeling teams.
dbt / SparkWarehouse patternsData qualityLineage
Life at Vegrade
How we work — and what work feels like
We're not a 500-person services factory. This is a focused engineering firm where how you work matters as much as what you ship.
Rhythm that respects focus
Core collaboration windows align with India business hours (IST), with flexibility when US or EU clients need overlap. We protect blocks for deep work—fewer standing meetings than enterprise IT shops.
Remote-first, outcome-based
Team members work from across India with intentional in-person syncs when it helps (kickoffs, design reviews). Success is measured by delivery and clarity, not hours logged in a specific office.
Learn on client-grade problems
You touch the same patterns we use on engagements: hybrid retrieval, policy-bound agents, MLOps gates, and procurement-friendly documentation—not disconnected take-home puzzles.
Psychological safety with high standards
We ask questions early, document assumptions, and review work in the open. Standards are high, but the goal is learning and shipping—not performative hustle.
Small team, visible impact
Headcount stays deliberately limited. Your commits, designs, and client readouts are seen by leadership—not buried under six layers of management.
Pairing and handoff culture
We teach what we build. Interns pair with staff engineers; full-time hires co-own runbooks and KT sessions so knowledge moves across the team.
What we look for
Signals that predict success here
Degrees help; shipped work and clear thinking matter more. We hire for judgment, communication, and reliability.
Strong fundamentals in TypeScript and/or Python — readable code and tests where they matter.
Comfort with ambiguity — you document assumptions and risks instead of hiding unknowns.
Genuine interest in data science, ML, LLM systems, agents, or retrieval — not slide decks about them.
Professional written English for client updates, RFCs, and runbooks.
For interns: a contiguous 10–14 week window. For full-time: core overlap with IST, flexibility for global client calls.
Hiring process
Straightforward, respectful of your time
- 01
Introduce yourself
Résumé or portfolio, role preference (intern vs full-time), and two sentences on what you want to learn or own.
- 02
Technical conversation
System design or code discussion at the level of the role—focused on tradeoffs, not trivia.
- 03
Working session
Short, bounded exercise for select candidates—only when mutually agreed, with clear time box.
- 04
Offer & onboarding
Written compensation, start date, mentor assignment, and expectations for your first 30 days.
Vegrade is an equal opportunity employer. We consider qualified applicants without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, disability, veteran status, or any other characteristic protected by law. If you need a reasonable accommodation during the process, mention it in your first email.
FAQ
Common questions
Ready to apply?
Work on AI systems that actually ship
Drop an email with your résumé, the role you're interested in, and two sentences on what you want to learn or own. We read every note.