Mentored, project-first and brutally practical. You ship code every day and build ten real-time projects that match what AI teams are hiring for right now.
Everyone follows the same 90-day plan and ships the same ten projects. Choose based on who pays and what you want out of it.
₹50,000for 3 months
Open enrolment for students and researchers who want the full programme with guaranteed mentorship.
Stipendstated in your offer
Selective. Clear a coding test and technical interview and get paid to build.
Per cohortafter a scoping call
For universities placing a batch, or companies upskilling teams on their own data.
| Fee-based | Paid | Sponsored | |
|---|---|---|---|
| Who pays | Intern, ₹50,000 | AlgoProfessor pays you | University or company |
| Selection | Application and orientation call | Coding test and interview | Agreement with institution |
| Projects | Ten programme projects | Ten projects plus live product work | Ten projects, domain-adapted |
| Mentorship | Group sessions, 10-day reviews | Embedded with R&D team | Cohort mentors, sponsor reports |
| Outcome | Certificate, portfolio, referrals | Experience letter, role consideration | Certificates, credit docs, trained team |
Pick a phase, then tap any day to see what you learn, the tools you use and what you submit. Gold days are the ten projects.
Tip: use the ← and → keys to move between days.
| Day | Topic | What you do | Tools |
|---|---|---|---|
| Phase 1: Foundations | |||
| Day 1 | Git, GitHub and dev setup | Git internals, branching, pull requests, merge conflicts, GitHub Actions and pre-commit hooks. Create the internship repository. | Git, GitHub CLI, VS Code / Cursor, uv, pre-commit |
| Day 2 | Modern Python I | Type hints, match/case, dataclasses, comprehensions, generators, itertools and functools. | Python 3.12, uv, ruff |
| Day 3 | Modern Python II | Pydantic v2 models, async/await, decorators, context managers and pytest fixtures. | Pydantic, asyncio, pytest, mypy |
| Day 4 | NumPy, Pandas and Polars | Vectorisation, broadcasting, groupby, joins, reshaping and Parquet I/O. Compare Pandas with Polars and DuckDB on a 5M-row file. | NumPy, Pandas, Polars, DuckDB |
| Day 5 | EDA and visualisation | Distributions, correlation, hypothesis tests and outliers. Interactive and publication-grade charts. | Plotly, Seaborn, SciPy, ydata-profiling |
| Day 6 | SQL and PostgreSQL | Joins, CTEs, window functions, indexing and EXPLAIN ANALYZE. ORM models and migrations. | PostgreSQL, SQLAlchemy 2, Alembic |
| Day 7 | Streaming data basics | Topics, producers, consumers and windowed aggregation. Stream a live public feed into a topic. | Redpanda / Kafka, aiokafka, FastStream |
| Day 8 | Classical machine learning | scikit-learn pipelines, cross-validation, metrics, Optuna tuning and SHAP explanations. | scikit-learn, Optuna, SHAP, joblib |
| Day 9 | Project 1: StreamSense | Build and deploy the real-time anomaly monitor. See the projects section for the brief. | Redpanda, scikit-learn, FastAPI, Plotly Dash |
| Day 10 | Boosting and feature engineering | XGBoost, LightGBM and CatBoost; lag, rolling and target-encoded features; stacking. | XGBoost, LightGBM, CatBoost |
| Day 11 | Neural networks from scratch | Forward pass, backpropagation and optimisers in NumPy. Verify gradients numerically. | NumPy |
| Day 12 | PyTorch 2 | nn.Module, DataLoader, mixed precision, torch.compile and experiment tracking. | PyTorch 2, Lightning, TensorBoard |
| Day 13 | CNNs and vision transformers | ResNet and ViT fine-tuning with augmentations and learning-rate schedules. | timm, Albumentations, torchvision |
| Day 14 | Detection, segmentation, tracking | Object detection, instance segmentation and multi-object tracking on video. | Ultralytics YOLO11, SAM 2, supervision |
| Day 15 | Model export and edge inference | ONNX export, INT8 quantisation and latency benchmarks on CPU. | ONNX Runtime, OpenVINO, TensorRT |
| Day 16 | NLP and transformers | Tokenisers, fine-tuning a BERT-family model for classification, evaluation. | Hugging Face transformers, datasets, evaluate |
| Day 17 | Embeddings and semantic search | Sentence embeddings, cosine similarity, HNSW indexing and multilingual models. | sentence-transformers, BGE-M3, FAISS |
| Day 18 | Project 2: VisionGuard | Build and deploy the real-time video safety system. See the projects section for the brief. | YOLO11, ONNX Runtime, WebSockets, FastAPI |
| Day 19 | NoSQL and caching | MongoDB aggregation, Redis data types, cache-aside patterns and task queues. | MongoDB, Redis, Celery |
| Day 20 | Vector databases | pgvector, Qdrant and Chroma; hybrid BM25 plus dense search; recall and latency comparison. | pgvector, Qdrant, Chroma |
| Day 21 | RAG foundations | Document parsing, chunking strategies, retrieval, reranking and grounded prompting. | Docling, PyMuPDF, LangChain, LlamaIndex |
| Day 22 | Advanced RAG | Query rewriting, HyDE, multi-query, contextual retrieval, parent-document retrieval and rerankers. | bge-reranker, Cohere Rerank, LlamaIndex |
| Day 23 | GraphRAG | Entity and relation extraction into a knowledge graph; graph plus vector retrieval. | Neo4j, LightRAG / GraphRAG, NetworkX |
| Day 24 | RAG evaluation | Golden datasets; faithfulness, answer relevancy and context precision; regression tests. | RAGAS, DeepEval |
| Day 25 | FastAPI and streaming APIs | REST, Server-Sent Events, WebSockets, async I/O and API-key auth. | FastAPI, Uvicorn, Pydantic |
| Day 26 | Docker and CI | Dockerfiles, docker compose, GitHub Actions running tests on every push. | Docker, Compose, GitHub Actions |
| Day 27 | Project 3: Enterprise Knowledge Navigator | Build and deploy the streaming GraphRAG assistant. See the projects section for the brief. | LangGraph, Qdrant, Neo4j, FastAPI, RAGAS |
| Day 28 | Front ends for AI apps | Streaming chat UIs, file upload, citations panel. Streamlit, Gradio and a basic Next.js chat. | Streamlit, Gradio, Next.js |
| Day 29 | Phase 1 review sprint | Refactor, raise test coverage above 70%, write docs site. | pytest-cov, mkdocs |
| Day 30 | Phase 1 panel review | Demo Projects 1–3 to the technical panel; mentor code review; personal gap plan. | All Phase 1 tools |
| Phase 2: LLM engineering | |||
| Day 31 | The 2026 LLM landscape | Frontier models (GPT-5 family, Claude 4.x, Gemini 3) versus open-weight (Llama 4, Qwen 3, DeepSeek-V3/R1, Mistral, Gemma 3). Reasoning models, cost and latency trade-offs. | LMArena, Artificial Analysis, OpenRouter |
| Day 32 | LLM APIs and SDKs | OpenAI Responses API, Anthropic Messages API, Gemini API; streaming, prompt caching, retries, cost tracking; one router over all three. | OpenAI SDK, Anthropic SDK, Google GenAI SDK, LiteLLM |
| Day 33 | Prompt and context engineering | Tokenisation and context windows; system prompts, few-shot, chain-of-thought, working with reasoning models; context engineering for long inputs. | tiktoken, DSPy, promptfoo |
| Day 34 | Structured outputs and tool calling | JSON schema outputs, function calling, validation and repair loops. | Pydantic AI, instructor, Outlines |
| Day 35 | Speech AI | Streaming speech-to-text, voice activity detection, text-to-speech; Indian-language models. | faster-whisper, AI4Bharat models, Pipecat, LiveKit Agents |
| Day 36 | Project 4: VoiceDesk | Build and deploy the real-time multilingual voice agent. See the projects section for the brief. | LiveKit / Pipecat, faster-whisper, LLM API, TTS |
| Day 37 | Serving open-weight LLMs | Local and server inference; batching, KV cache, throughput and latency benchmarks. | Ollama, vLLM, SGLang, llama.cpp |
| Day 38 | Quantisation | GGUF, AWQ, GPTQ and FP8; measuring quality loss against speed and memory. | llama.cpp, AutoAWQ, bitsandbytes |
| Day 39 | Fine-tuning foundations | LoRA and QLoRA, chat templates, dataset curation and synthetic data generation. | PEFT, Unsloth, datasets |
| Day 40 | Supervised fine-tuning | Fine-tune a Qwen 3 or Llama model on a domain dataset with experiment tracking. | TRL SFTTrainer, Axolotl, Weights & Biases |
| Day 41 | Preference and RL tuning | DPO and ORPO for preferences; GRPO with verifiable rewards for reasoning tasks. | TRL DPOTrainer / GRPOTrainer |
| Day 42 | Distillation and small models | Distil a frontier model's outputs into a small model; measure cost per answer. | TRL, vLLM |
| Day 43 | LLM evaluation | Benchmarks, LLM-as-judge, custom task evals and regression suites. | lm-evaluation-harness, DeepEval, promptfoo |
| Day 44 | RAG versus fine-tuning | Decision framework; test RAG, fine-tuning and hybrid on the same task. | RAGAS, W&B |
| Day 45 | Project 5: DomainLLM | Fine-tune, quantise and serve a domain model. See the projects section for the brief. | Unsloth, TRL, vLLM, FastAPI |
| Day 46 | Multimodal models | Vision-language models for image and document reasoning; grounding and OCR limits. | Qwen-VL, Gemini, GPT vision, Claude vision, CLIP |
| Day 47 | Document intelligence | Layout parsing, tables, forms and invoices into structured JSON. | Docling, PaddleOCR, Pydantic |
| Day 48 | Image generation basics | Diffusion models, prompting, ControlNet and responsible use. | diffusers, FLUX / SDXL |
| Day 49 | Safety and guardrails | Prompt injection, jailbreaks, PII redaction and safety classifiers; red-team your own app. | Llama Guard, NeMo Guardrails, Presidio |
| Day 50 | Observability | Tracing, token cost, latency and quality dashboards for LLM apps. | Langfuse, LangSmith, Arize Phoenix, OpenTelemetry |
| Day 51 | Responsible AI and data law | DPDP Act 2023, consent, data minimisation, bias checks and explainability for health data. | Fairlearn, SHAP, checklists |
| Day 52 | LLM gateways and API design | OpenAI-compatible APIs, rate limits, semantic caching and model fallbacks. | LiteLLM Proxy, FastAPI, Redis |
| Day 53 | Cloud AI platforms | Deploy one model endpoint on a managed platform; compare cost and quotas. | AWS Bedrock / SageMaker, Vertex AI, Azure AI Foundry |
| Day 54 | Project 6: MedLens | Build and deploy the multimodal clinical assistant. See the projects section for the brief. | VLM, RAG, Presidio, NeMo Guardrails, Streamlit |
| Day 55 | Research methods | Reading and reproducing papers, experiment design, tracking and ablations. Paper co-authoring track begins. | arXiv, Papers with Code, W&B |
| Day 56 | R&D sprint I | Reproduce a result from a recent paper on your own hardware budget. | Paper's code, W&B |
| Day 57 | R&D sprint II | Ablations, error analysis and an honest write-up of what did not work. | W&B, Matplotlib |
| Day 58 | Technical writing | Turn one project into an IEEE-format short paper or a technical blog post. | LaTeX / Overleaf |
| Day 59 | Phase 2 review sprint | Refactor, add evals to CI, update docs for Projects 4–6. | pytest, promptfoo, GitHub Actions |
| Day 60 | Phase 2 panel review | Demo Projects 4–6 to senior engineers; code review; gap plan. | All Phase 2 tools |
| Phase 3: Agentic AI | |||
| Day 61 | Agent foundations | ReAct, plan-and-execute and reflection. Build a tool-using agent loop with no framework. | Python, LLM API |
| Day 62 | Model Context Protocol (MCP) | Build MCP servers exposing tools, resources and prompts; transports and auth. | MCP Python SDK, FastMCP |
| Day 63 | Project 7: OpsPilot | Build and deploy the MCP-powered operations assistant. See the projects section for the brief. | MCP, FastMCP, PostgreSQL, GitHub API |
| Day 64 | LangGraph | State graphs, conditional edges, checkpoints, interrupts and human approval steps. | LangGraph, LangGraph Studio |
| Day 65 | Agent SDKs compared | Handoffs, guardrails and tracing across three SDKs; same task, three builds. | OpenAI Agents SDK, Claude Agent SDK, Google ADK |
| Day 66 | CrewAI | Roles, tasks, crews and flows for business workflows. | CrewAI |
| Day 67 | Multi-agent patterns | Supervisor, swarm, hierarchical and group-chat patterns; when each fails. | AutoGen / AG2, LangGraph |
| Day 68 | Agent memory | Short-term, episodic and semantic memory; memory write and recall policies. | Mem0, Zep, Letta |
| Day 69 | Coding agents | Sandboxed code execution, test-driven repair loops, SWE-bench style tasks. | E2B / Docker sandboxes, pytest |
| Day 70 | Deep research agents | Planner, searcher, writer and editor agents with source citations. | Tavily / Exa, LangGraph |
| Day 71 | Agent-to-agent communication | A2A protocol, event-driven agents and message queues. | A2A SDK, Redis Streams, RabbitMQ |
| Day 72 | Project 8: DevSquad | Build and deploy the multi-agent engineering team. See the projects section for the brief. | CrewAI / LangGraph, Docker, GitHub API |
| Day 73 | Browser and computer-use agents | Web navigation, form filling and data extraction from plain-English instructions. | Playwright, Browser Use, Stagehand |
| Day 74 | Agent security | Tool-borne prompt injection, least privilege, OAuth for MCP, sandboxing and secret handling. | OWASP LLM Top 10, OAuth 2.1 |
| Day 75 | Agent evaluation | Trajectory and outcome evals, tau-bench and GAIA-style tasks, custom harness. | LangSmith, Langfuse evals, DeepEval |
| Day 76 | Production hardening | Timeouts, retries, circuit breakers, fallbacks and per-task cost budgets. | Tenacity, LiteLLM |
| Day 77 | Real-time voice and vision agents | Low-latency speech-to-speech agents with tool use and screen or camera input. | Realtime APIs, LiveKit Agents |
| Day 78 | Durable agent workflows | Long-running workflows that survive crashes, with scheduled and triggered runs. | Temporal, n8n |
| Day 79 | Agent UX | Streaming UIs, approval steps, generative UI and showing the agent's work. | AG-UI, CopilotKit, Next.js |
| Day 80 | Deploying agents | Containers, background workers, queues and autoscaling for agent services. | Docker, Cloud Run / Railway, Redis |
| Day 81 | Project 9: SidekickBrowser | Build and deploy the browser co-pilot. See the projects section for the brief. | LangGraph, Playwright, FastAPI, Chrome extension |
| Day 82 | Kubernetes for AI | Pods, services, Helm charts and GPU scheduling basics. | Kubernetes, Helm, kind / k3s |
| Day 83 | LLMOps and CI/CD | Prompt and model versioning, eval-gated releases and canary deployments. | GitHub Actions, MLflow, promptfoo |
| Day 84 | Monitoring | Metrics, logs, traces and alerts for agent systems. | Prometheus, Grafana, OpenTelemetry |
| Day 85 | Capstone build I | Architecture and MCP servers: market data, news, fundamentals, orders, risk, notifications. | MCP, FastAPI, PostgreSQL |
| Day 86 | Capstone build II | Analyst, quant, risk and execution agents orchestrated as a graph. | LangGraph / AutoGen |
| Day 87 | Capstone build III | Risk rules, backtesting and paper-trading integration; kill switch. | Backtrader / vectorbt, broker paper API |
| Day 88 | Capstone build IV | Deploy on Kubernetes with dashboards; load and failure tests. | Kubernetes, Grafana |
| Day 89 | Docs, demo and portfolio | README, API docs, architecture diagrams, 5-minute demo video, portfolio and LinkedIn update. | mkdocs, Swagger, Excalidraw |
| Day 90 | Project 10: AlphaTrader demo day | Twenty-minute live demo of the capstone to the CEO panel, then Q&A. | All Phase 3 tools |
Ten core projects, one every nine days, plus twelve bonus projects you choose by domain. Each is deployed, public on your GitHub and mapped to a job title.
Scores a live sensor stream every second and flags anomalies as they happen.
Streams public air-quality or IoT sensor readings through a message broker, runs an anomaly model online and alerts on a live dashboard with drill-down history.
Detects missing helmets and vests on a live camera feed at 20+ FPS on CPU.
YOLO11 detection with tracking on RTSP or webcam video, exported to ONNX for edge devices, with a WebSocket alert feed and a daily incident report.
Streams cited answers token by token and re-indexes documents as they change.
GraphRAG assistant over 50+ company documents: hybrid retrieval, reranking and a knowledge graph, with page-level citations and a live RAGAS quality dashboard.
Answers callers in Telugu, Hindi or English with under one second of response latency.
A voice agent that books appointments or answers FAQs by phone or browser, calls tools for real bookings and hands off to a human when unsure.
Serves a fine-tuned small model with streaming responses and a public benchmark against frontier APIs.
Curate a domain dataset, fine-tune with QLoRA and GRPO, quantise, and serve with vLLM. Publish accuracy, latency and cost per 1,000 answers against a frontier model.
Reads an uploaded scan and report and returns a grounded draft summary in seconds.
A clinician-assist tool on public datasets: a vision-language model reads chest X-rays or retinal images, RAG grounds the summary in guidelines, and guardrails redact personal data and block unsafe advice.
Acts on live company systems through MCP, with a human approval step before any write.
An assistant connected to a database, GitHub and email/calendar through your own MCP servers. It answers questions, drafts actions and executes them once approved.
Picks up a new GitHub issue and opens a tested pull request without human coding.
Planner, coder, reviewer and tester agents working in isolated Docker sandboxes, running tests until they pass and explaining every change in the PR.
Carries out multi-step web tasks live in your browser from one English instruction.
A Chrome extension backed by a LangGraph agent that navigates sites, fills forms and extracts structured data, pausing for confirmation before payments or submissions.
Four agents trade NIFTY 50 in paper mode on live market data, with a risk agent that can halt the floor.
The grand capstone: analyst, quant, risk and execution agents coordinated over six MCP servers, deployed on Kubernetes with Grafana monitoring and a full audit trail. Paper trading only.
Build at least two during the R&D days (55–58) or alongside Phase 3. Great for Ph.D. and post-doc interns who want work close to their research.
Farmer photographs a crop leaf; the app detects disease and replies with a Telugu or Hindi voice note.
After Phase 2
RAG over Indian bare acts and judgments that answers with verified section and case citations.
After Phase 1
Extracts GST invoices from PDFs and emails, then an agent reconciles them against ledger entries live.
After Phase 2
Tracks brand sentiment across news and social posts in English and Indian languages, with spike alerts.
After Phase 2
Generates question papers mapped to Bloom's levels and course outcomes, then auto-grades answers with rubrics.
After Phase 2
Scores URLs and emails for phishing in real time and explains the verdict in plain language.
After Phase 1
Counts vehicles per lane from junction CCTV and suggests signal timings as traffic changes.
After Phase 1
Estimates remaining useful life from vibration sensors; an agent raises the maintenance work order.
After Phase 3
Live call transcription with sentiment, compliance flags and an auto-written call summary.
After Phase 2
Skin-lesion triage from phone photos with confidence scores, explanations and a see-a-doctor threshold.
After Phase 2
An agent that reads an arXiv paper, writes the code and reproduces its main table in a notebook.
After Phase 3
A simulated robot arm that picks objects from spoken commands using a vision-language-action pipeline.
After Phase 3
Hiring has moved from model theory to shipped systems: retrieval, fine-tuning, agents, evaluation and deployment.
The baseline every AI job description assumes.
Python 3.12uvGitSQLDockerFastAPIasync I/OpytestStill the basis for vision, forecasting and tabular work.
scikit-learnXGBoostPyTorch 2ViTYOLO11SAM 2ONNXUsing frontier and reasoning models well, cheaply and reliably.
Context engineeringStructured outputsTool callingPrompt cachingLiteLLMDSPyThe most common production GenAI pattern in Indian enterprises.
Hybrid searchRerankingGraphRAGpgvectorQdrantRAGASOwning a model when APIs are too costly or data cannot leave.
LoRA / QLoRADPOGRPOUnslothTRLvLLMGGUF / AWQThe fastest-growing and best-paid specialisation this year.
MCPA2ALangGraphOpenAI Agents SDKClaude Agent SDKGoogle ADKCrewAIAutoGenWhat separates a demo from a system a company will run.
LangfuseLangSmithOpenTelemetryKubernetesHelmGitHub ActionsEvery project here runs on live data or live users.
Kafka / RedpandaWebSocketsSSEStreaming STT/TTSLiveKitRequired for healthcare, finance and government work.
DPDP Act 2023PII redactionPrompt-injection defenceRed-teamingGuardrailsFreshers with a working GenAI or agent project, not just a certificate, are the ones negotiating above the standard entry band.
Indicative ranges compiled from public 2026 salary reports (Glassdoor and Indeed summaries, CloudThat, Taggd, Masai, SCDL, 360 Digital Transformation, Amquest; reviewed September 2026). Rows marked * are AlgoProfessor estimates derived from those reports. Pay is highest in Bengaluru and Hyderabad and at product companies and GCCs. Not an offer or placement guarantee.
Last paid batch: 100 started, 2 finished all 90 days, both placed. Consistency is the whole difference.
90–100: excellence certificate, recommendation letter, LinkedIn endorsement. 75–89: completion certificate and endorsement. 60–74: participation certificate. Below 60: mandatory review; the internship may end.
The fee-based internship is ₹50,000 for 3 months. The paid internship has no fee; selected interns receive a stipend stated in their offer letter. University or company sponsored cohorts are quoted per cohort after a scoping call.
B.Tech, M.Tech, Ph.D. scholars and post-doctoral researchers from any branch. Phase 1 starts from Git and Python, so prior AI experience is not required for the fee-based model. The paid internship needs a public GitHub profile with real work.
The programme runs from Hyderabad and online. Sponsored cohorts can be delivered on campus, online or hybrid.
Ten core real-time projects, one every nine days, ending with a multi-agent trading capstone, plus at least two of twelve bonus projects chosen by domain. Every project is deployed and public on your GitHub.
You push code and documentation to your internship repository by 5:00 PM every day with the commit format [Day-N] and email the link. Timeliness and commit quality count for 10 of the 100 marks.
Scores of 90–100 earn a certificate of excellence, letter of recommendation and LinkedIn endorsement; 75–89 a completion certificate and endorsement; 60–74 a participation certificate.
Sponsored cohorts get documentation support for NEP 2020 and AICTE internship credits, plus progress dashboards and phase reports.
Fill in the application form on this page or email cto@algoprofessor.com with your CV, a short statement of interest and your GitHub URL.
Takes about two minutes. It goes straight to the AlgoProfessor technology office.
Prefer email? Write to cto@algoprofessor.com with your CV and GitHub URL.