AI Engineer Intern

Location: Bengaluru - JP Nagar (On-site) Working: 6 days (we're locked in and expect same from you) Duration: 6 months Stipend: ₹35,000 – ₹60,000 per month [JD on Notion](https://app.notion.com/p/AI-Engineer-Intern-404aa6925a7b808a990efe7ba149bc03) **About the Role** We're a VC-funded startup building at the intersection of AI and education, and we crossed 1 million+ users in our first 6 months. We use generative AI to turn Govt exam prep into cinematic, watchable content. Every video is produced by AI, from script to final frame. This is an engineer's role first. You'll write Python, build AI agents, and ship the tools that power our video production pipeline. You'll also need to know the craft of AI video generation end to end, and have the eye to tell a great shot from a broken one. We don't want someone who only prompts. We don't want someone who only codes and has never touched a generation model. We want the rare person who does both. Ideally, you've already built end-to-end AI agents for real problems, generated video and audio with AI yourself, and can tell frame by frame what's working and what isn't. What You'll Do **AI Agents & Automation**: Design and build end-to-end AI agents and automated pipelines in Python that run the production workflow: script → assets → assembly → QA. This is the core of the role. **Generative Video & Media**: Build cinematic videos with models like Google Veo, Higgsfield, Kling, Runway and Grok Video. Turn narration and audio into synced, animated video. Generate images and animate them into motion. **Pipeline & Tooling**: Build the internal tools that make production faster and more consistent: prompt libraries, asset pipelines, ComfyUI workflows, API integrations and light automation. When you see a manual process, build the thing that removes those steps. **Review & Quality**: Review generated output critically. Judge which frames and shots are good and diagnose failure modes such as drift, artifacts, inconsistency, broken lip-sync and warping. Feed what you learn back into prompts, models and workflows. Knowing why a generation failed matters as much as making one that works. **Scripting & Storyboarding**: Work across the full creative pipeline, turning exam concepts into scripts, storyboards and visual narratives that actually land with learners. **End-to-End Ownership**: Take a piece of tooling or content from "here's a vague idea" to "it's live in front of users." Over time, handle more of the work end to end with less hand-holding. A pipeline that only works on your laptop is worth zero. **The Pipeline You'll Live Inside** Stage What "good" means Script Accurate exam content, tight pacing, written to be watched rather than read Storyboard Every shot has a purpose, and the visual narrative matches the concept Image Generation On-style, on-subject, consistent characters and look Video Generation Cinematic motion with no warping, drift or artifacts Audio & Sync Clean narration (ElevenLabs or similar) and correct lip-sync and timing Assembly & QA Automated checks catch broken output before a human has to Tools & models you'll touch: Veo · Kling · Runway · Higgsfield · Grok Video · ElevenLabs · ComfyUI · LoRAs · ControlNet · LangChain / LlamaIndex · n8n · Claude Code / Codex · RunPod / AWS · Docker **What You Bring** Python (mandatory): You code, and you've built real things with it, not just notebooks. This is a building role. You've built AI agents (mandatory): You've shipped end-to-end AI agents or tools that solve actual use cases. Show us what you built, how it works, and what broke along the way. Hands-on AI media experience (mandatory): You've actually generated video and/or audio with models like Veo, Kling, Runway, Higgsfield, ElevenLabs or similar. We mean you've built with them, not just read about them. Full-Pipeline Understanding: Scripting → storyboarding → image generation → video generation → review. You know what good looks like at each stage and can spot what's broken. A Critical Eye: You can tell cinematic output from AI slop. You can explain why a shot isn't working and what would fix it. Model Fluency: You're familiar with video, image and audio generation tools, and ideally with diffusion workflows (ComfyUI, LoRAs, ControlNet). Proof You Build Things: A GitHub you're proud of, deployed tools or agents, a reel or portfolio of AI-generated media, side projects or hackathons. We care more about what you've built than where you studied. Current students, recent grads and self-taught builders are all welcome. **Nice to Have** Experience fine-tuning models, training LoRAs, or building custom ComfyUI / diffusion workflows. Familiarity with agent frameworks (LangChain, LlamaIndex or similar), API orchestration, and automation tools (n8n, Claude Code, Codex). Character and style consistency techniques, such as LoRAs, reference libraries, and keeping a look coherent across shots. Comfort running models on cloud GPUs (RunPod, AWS) and containerizing workflows (Docker). Traditional editing or motion skills (Premiere, After Effects, DaVinci, CapCut), which help you know what "finished" looks like. An understanding of Bharat / Tier 2-3 learners and Hindi-first audiences, who watch on a phone in short bursts. The kind of curiosity that has you tracking new model releases the week they drop, because you actually find it interesting. **Culture Fit** Adaptive to Ambiguity: You keep moving even when the brief or the technical path isn't fully defined, and you ask good questions to fill the gaps. Builder & Problem Solver: When an agent misbehaves or a generation won't cooperate, you stay with it. You read docs, test approaches, and break the problem down until it works. Self-Starter & Fast Learner: You don't wait to be told every step. In a space moving this fast, half the job is figuring out what nobody has documented yet. AI-Native Builder: You spend real time on design and planning, then lean on AI tools to ship fast. You have the taste to keep quality high and the guardrails (tests, evals) to keep it reliable. **What You Get** Real ownership from week one. You ship real systems and real content to real users. Direct mentorship from the founding team. You get pairing, code and output reviews, and honest feedback. We want you to leave a meaningfully stronger builder than when you joined. Early access to the newest generation models and workflows. Portfolio-worthy work you can actually show. Certificate, letter of recommendation, and a PPO for strong performers.