MargdarshakAI helps students become placement-ready and gives companies a direct view of student profiles and progress. It is built by Get Set Skilled™ Private Limited, a DPIIT-recognized startup incubated at GL Bajaj Center for Research and Incubation (GLBCRI), Greater Noida.
About the Role
You will build the AI that powers both sides of hiring on MargdarshakAI, students and recruiters, working directly with the founders and our engineering team on systems that run in production.
Key Responsibilities
- Build ranking and matching models that score candidate profiles against job requirements.
- Build LLM pipelines that generate structured content (questions, test cases, learning paths) and validate it automatically.
- Build voice AI pipelines with speech-to-text, text-to-speech and real-time streaming.
- Design rubric-based and LLM-as-a-judge scoring for text, spoken and code responses, and turn results into explainable reports.
- Build guardrails against prompt injection, PII leakage, toxic output and hallucination, and red-team AI features.
- Maintain eval suites and regression tests that run before every prompt or model change.
- Support AI governance: versioning, audit logs, bias checks, human review and DPDP Act, 2023 compliance.
- Build adaptive RAG for chat (query routing, query rewriting, hybrid search, reranking and self-correcting retrieval) on top of vector search, and expose models as FastAPI endpoints for the Next.js app.
- Track accuracy, latency and cost, and document experiments.
Required Technical Skills
- ML & NLP: Python, pandas, scikit-learn, PyTorch, Hugging Face Transformers, spaCy; embeddings, NER and ranking metrics (precision@k, NDCG).
- LLMs: Gemini, OpenAI or Claude APIs; structured outputs and tool calling; prompt design; RAG and adaptive RAG (query routing, hybrid search, reranking, corrective or self-RAG).
- Security & Evals: OWASP Top 10 for LLM Applications; guardrail tools (Guardrails AI, NeMo Guardrails, Llama Guard, Presidio); eval tools (Promptfoo, RAGAS, DeepEval); LLM tracing (Langfuse or LangSmith).
- Speech: Whisper or cloud speech-to-text / text-to-speech APIs; WebSockets.
- Backend: FastAPI, Docker, Celery or Redis queues, Git, basic AWS/GCP/Azure.
- Data: PostgreSQL, pgvector or ChromaDB/Pinecone.
Good to Have
- LangChain, LangGraph or LlamaIndex for agentic and multi-step retrieval workflows.
- A voice bot, chatbot or AI agent built end to end.
- Code execution sandboxes (Judge0, Docker-based runners).
- Red-teaming LLM apps, AI security CTFs, or awareness of NIST AI RMF / ISO/IEC 42001.
- Hindi or bilingual (Hindi-English) text and speech.
- A GitHub profile, Kaggle work or deployed project we can look at.
What You Will Learn and Gain
- Ship end-to-end AI features used by real students and recruiters.
- Learn production LLM engineering: adaptive RAG, evals, guardrails, AI security and voice AI.
- Exposure to AI system design and product thinking.
- Internship certificate and letter of recommendation on successful completion.
Work Setup and Eligibility
- Hours: 9:00 AM to 6:00 PM, Monday to Friday from office; 2nd, 4th and 5th Saturdays work from home.
- Reviews: Weekly goals with progress reviews.
- Eligibility: Final-year students or recent graduates (B.Tech/B.E., MCA, M.Tech, M.Sc.) in CS, IT, AI/ML, Data Science or related fields.
- Keep your resume (PDF) ready for upload.
- Spend at least 10 minutes exploring https://margdarshakai.com before answering the product-related questions.
- Ensure all information provided is accurate.
Stipend & PPO:
Location: Knowledge Park-3, Greater Noida (Work From Office)Duration: 3 monthsPre-Placement Offer: Up to ₹6 LPA, based on performance
Please Note:
- We strongly discourage the use of AI tools to generate responses, as we value original thinking over polished answers. Generic or AI-generated responses may be shortlisted lower than thoughtful and authentic submissions.
- We carefully review every application and appreciate the time and effort you invest in applying.
We look forward to learning more about you. Best of luck!
This position is open to registered MargdarshakAI users.