DreamPi: AI Language Engineer & QA Intern (Indic Languages)

Company: DreamPi Role: AI Language Engineer & QA Intern, Indic Languages Location: Chennai, Tamil Nadu (on-site) Type: Internship, full-time Duration: 2 months, with a pre-placement offer (PPO) for strong performers Reports to: Founding team, DreamPi

About DreamPi

DreamPi builds AI for Bharat. Most frontier AI today is trained, tested and tuned in English, which means it breaks the moment a real user in Indore, Mysuru, Warangal or Kochi types the way they actually type. We are building multi-modal AI systems that work in Indic languages the way they are genuinely spoken and written, not the way a translation layer imagines them.

We are assembling a small in-house team of native language experts who sit next to the engineers building the models. This is the first cohort.

The Role

You will be responsible for two things at once: helping build DreamPi's Indic language AI, and running QA on the AI infrastructure that serves it.

This is not an annotation job. You will be the person who decides whether a model's Hindi is actually good Hindi, and who then goes and fixes it.

What you'll do

Build

- Design, write and curate high-quality evaluation and training data in your native language(s)
- Build test suites that stress model behaviour across dialect, register, script, code-mixing (Hinglish, Kanglish, etc.), transliteration and honorifics
- Work with the engineering team on prompt design, instruction tuning data and model behaviour specs for Indic use cases
- Translate and localise product surfaces, but as a linguist rather than a dictionary

QA the AI

- Run structured, repeatable QA cycles across our models and AI infrastructure, covering text, speech and multi-modal
- Record, reproduce and document failures: hallucination, script errors, wrong grammatical gender, broken transliteration, tone mismatch, cultural inaccuracy, unsafe output
- Score and rank model outputs against defined rubrics; maintain the evaluation scorecards
- Regression-test releases: catch what broke between versions, before users do




- Own the feedback loop back into the model. File it, explain it, follow it until it's fixed

What the day actually looks like Testing many model versions across several languages. Writing down precisely why an output is wrong in a way an engineer who doesn't speak that language can act on. Building the evaluation set that didn't exist yesterday. Sitting in the room while the model gets better because of what you found.

Languages We're Hiring For

We are hiring across eight Indian languages . All eight are equally required, and we need coverage across every one of them.

Hindi Devanagari Kannada Kannada Telugu Telugu Malayalam Malayalam Marathi Devanagari Bengali Bengali Gujarati Gujarati Punjabi Gurmukhi

You need to be native-level in at least one of these eight.

If you cover more than one, that shapes your role. Multi-language candidates get wider scope, more ownership, and a stronger case at PPO time. We are actively looking for people who work across several of these.

Language bar: we are strict about this

- You must be a native-level reader and writer of at least one of the eight
- Native script fluency is mandatory , whether that's Devanagari, Kannada, Telugu, Malayalam, Bengali, Gujarati or Gurmukhi. Romanised-only proficiency will not explicit our screening
- You should be able to spot the difference between a sentence that is grammatically correct and one that a real speaker would actually say
- Every shortlisted candidate takes a written language proficiency test, in every language they claim. There is no way around it

Who We're Looking For

Required

- Native-level written and spoken fluency in at least one of: Hindi, Kannada, Telugu, Malayalam, Marathi, Bengali, Gujarati, Punjabi
- Genuine interest in AI.



You've used LLMs seriously, you've noticed where they fail, and it bothers you
- Strong written English (this is our working language for documentation and bug reports)
- Precision and patience. QA rewards people who notice the small wrong thing on the two-hundredth example
- Based in Chennai, or able to relocate, and comfortable working from our office

Nice to have

- A second or third language from the eight
- Any background in linguistics, literature, translation, journalism, or language teaching
- Basic Python, or comfort with spreadsheets, scripts and structured data
- Familiarity with prompt engineering, model evaluation, or annotation tooling
- Prior exposure to speech data, ASR/TTS, or transcription work

Open to: students, recent graduates, career-switchers and language professionals moving into AI. We care about the language and the judgment, not the degree.

What You Get

- Direct, hands-on work on frontier Indic language AI. This is a core function, not a support one
- A seat with the founding team; your findings change what ships
- Mentorship on AI evaluation, model behaviour and applied NLP
- Internship certificate and letter of recommendation
- PPO consideration for a full time role at the end of the 2 months
- Stipend, discussed at interview, with scope for candidates covering multiple languages

Hiring Process

1. Apply with your languages clearly listed
2. Language proficiency test , written, in your claimed native script
3. AI reasoning round , a practical exercise: evaluate real model outputs in your language and tell us what's wrong
4. Conversation with the founding team , on-site in Chennai
5. Offer

How to Apply Apply via LinkedIn, or email [email protected] with:

- Your CV
- Subject line: AI Language Intern: [Your Languages] (for example, AI Language Intern: Telugu, Kannada)
- A short paragraph written in your native script , telling us why Indic AI matters to you

DreamPi is an equal opportunity employer. We hire on language ability, judgment and curiosity.

📌 AI Language Engineer & QA Intern (Chennai)
🏢 DreamPi
📍 Chennai