PyCon India 2025
who it's for
python developers, machine learning engineers, ai infrastructure engineers, backend engineers, devops engineers, data engineers
topics
Celery with Kubernetes; asyncio and socket multiplexing; compiling Python to WebAssembly; LoRA and subspace fine tuning; energy-efficient LLM serving with vLLM; RAGs with async scraping and re-ranking; Pydantic prompt validation; Locust + Prometheus + Grafana; Ruff linting with Dagger deploy; DuckDB for analytics; MCP server architectures (Rube); TinyML with MicroPython
| title | author | length |
|---|---|---|
| Fellowship of the Stack: Scientific Discovery with Python | Dawn Wages | 50m |
A practical tour of scientific Python that names tools, tradeoffs and numbers: a Python packaging survey has 16,000 respondents and closes end of December… | ||
| Celery and Kubernetes for a Fast, Scalable, and Robust Workflow Orchestration | Param Rajani | 34m |
Describes a production migration from AWS Lambda and Step Functions to Celery Canvas with Kubernetes that replaces state transitions and per-invocation costs… | ||
| Python and Music: Building a Music Tutor | Lakshya Gupta, Anant Gupta | 22m |
A Python pipeline converts uploaded MP3/WAV into separated tracks using Demucs or UVR, converts melody to MIDI with BasicPitch/PrettyMIDI, and builds graded… | ||
| Lessons From The Trenches: Building Rube | Jayesh Sharma | 31m |
Rube is a universal MCP server that connects LLMs to 500+ apps and manages huge context by writing bulky tool responses to a machine sandbox instead of… | ||
| Compute-Scaling Methods at Inference Time | Rutvik Acharya, Nitin Agarwal | 27m |
Compares specific inference-time methods and their tradeoffs using a quantized Llama 3 2.1B 4-bit model on Colab with 50 reasoning prompts, evaluated by… | ||
| RapidDoc - A Rule-Based Fine-Tuned Model for Enforcing Style Guide Rules | Gaurav Trivedi | 23m |
An open‑source pipeline fine‑tunes flan-ts-base on a premium dataset of rule-specific examples to flag and rewrite violations across 150+ style rules with… | ||
| Python Sockets at Scale: I/O Multiplexing and Asyncio | Rohan Reddy Alleti | 28m |
Shows that per-connection threads hit the GIL so they run concurrently not parallel, then demos using the OS readiness syscalls (select/poll/epoll/kqueue) and… | ||
| Reviving, Modernizing, and Packaging Unmaintained Code | Ajinkya P. Dahale | 24m |
Revives a 2016-era desktop app (Osdag) and makes it installable again after 4–5 years of stasis by fixing relative imports, adding setuptools/conda entry… | ||
| Artificial Information: How Today's AI is Changing Information | Katharine | 39m |
Argues that current AI turns murky, poorly governed datasets into proclaimed facts — cites ICE linking disparate records to deport people and Google Gemini’s… | ||
| Mastering Prompts with Feedback and Pydantic | Mahima Arora, Aarti Jha | 30m |
A productionized prompt optimization pipeline uses LangGraph for orchestration, Olama for local model runtime, Streamlit for UI and Pydantic validation as a… | ||
| From Stress to Success: Load Testing Python Apps & Visualizing Performance | Allen Y | 31m |
Hands-on walkthrough of building a load-testing pipeline that actually stores and visualizes results: Locust + Prometheus + Grafana with a Locust exporter and… | ||
| Compiling Python to WASM | Kumar Anirudha, Farhaan Bukhsh | 34m |
Compares existing approaches (Pyodide, PyScript, RustPython, Py2WASM, SPy, MicroPython) and shows a new native compiler called Waspy that takes Python source… | ||
| Memory Management in Python: Foundations, Problems, and Modern Techniques | Jithu Sunny | 31m |
Explains CPython’s two-phase memory reclamation: reference counting plus generational GC with gen0 triggered after ~700 allocations and gen1 after ten gen0… | ||
| Lint Like Lightning, Deploy Like a Ninja: Ruff + Dagger in Action | Urvashi Choubey | 30m |
Ruff replaces flake8, black and isort with a Rust engine that runs up to 100x faster and lints the OpenStack Nova repo (≈1,400 files) in about 0.03s, giving… | ||
| Implementing an MCP Server for DBMS in Python — YDB’s Experience | Ivan Blinkov | 33m |
YDB implements a Python-based MCP server that exposes SQL access for LLMs, using Python-based MCP server with a local stdio transport so the server reuses the… | ||
| Navigating Real-World Challenges in a Production-Grade Multi-Agent System | Sibin Bhaskaran | 30m |
A production conversational assistant for cricket data often fails on specifics: an observed Agent Kohli failure answers ODI World Cups with IPL data. The… | ||
| RAGs to Riches: Efficient WebChatbot with Async Scraping & Hybrid Re-Ranking | Charan, Srikanth | 29m |
A production RAG pipeline that scrapes a 900+ page site and drops full ingestion from 5h18m to 40m using asyncio 5h18m→40m and incremental SHA‑256 content… | ||
| Is FastAPI really fast ? | Princekumar Dobariya | 31m |
Benchmarks compare FastAPI (async vs sync) to Flask, Django, Express and Go Gin using WRK and TechEmpower suites, run on Dockerized EC2 4 vCPU / 8 GB with 60s… | ||
| Geometry of Efficient Fine Tuning: LoRA, Intrinsic Dimension, Subspace Learning | Preethi Srinivasan | 33m |
Reproduces MNIST intrinsic-dimension results on a 784-200-200-10 MLP and shows D90 = 750 from ~200k parameters, meaning 750 learned directions hit ~90% of best… | ||
| Boundaries of Multi Agent Systems | Robin Gupta | 30m |
An open-source testing agent called TestZeus Hercules executes plain-English tests end-to-end using a multi-agent architecture: a centralized planner agent… | ||
| From Scripts to Solvers: Building Agentic AI Systems in Python That Think and Do | Vatsal Shah | 27m |
Argues that replacing brittle scripts with model-driven agents yields faster prototyping and fewer bugs, citing a measured 10x faster feature cycle at AWS… | ||
| Understanding Plugin Architecture for Python Packages | Ashish Shukla | 30m |
Explains how Python packages advertise and load plugins using importlib.metadata entry points, pyproject.toml groups and a simple registry loader that calls… | ||
| Engineering Custom Diffusion Workflows with ComfyUI: A Developer's Guide | Shreya, Siddhi Kommuri | 30m |
ComfyUI is a visual, node-based tool that lets developers build reproducible Stable Diffusion pipelines by wiring node-based Python nodes (examples as short as… | ||
| The Anatomy of Django Migrations | Bhuvnesh Sharma | 26m |
Explains exactly how Django turns model edits into applied schema changes by using ModelState/AppState/ProjectState snapshots, an autodetector and questioner… | ||
| Hope, Hesitation and Hello World' : A Girl's Path from Confusion to Clarity | Heer Thakkar | 24m |
Describes a concrete learning path: moves from copy-paste to coding by switching from endless YouTube tutorials to hands-on practice, uses ChatGPT for targeted… | ||
| Finding a Needle in the Haystack - Debugging Performance Issues | Puneet Khushwani | 32m |
Practical debugging steps for Sev1 slowness using Python profilers with concrete results: deterministic profilers (cProfile) and sampling tools (py-spy… | ||
| Day 1 Closing Address | Bhavin Gandhi | 08m |
Conference reports 1100+ attendees and 40+ sessions, with six additional talks squeezed into the schedule; keynotes named Shakthi and Don headline the day… | ||
| Day 2 Opening Address | Bhavin Gandhi | 08m |
Conference celebrates its 16th edition since 2009 and runs workshops on the 12th with the main conference across yesterday and today and Dev Sprints tomorrow… | ||
| Know Thy Packets: Demystifying Network Fundamentals with Scapy | Shrayas, Rengaraj D | 34m |
Shows a real curl to icanhazip that uses ten packets for HTTP and thirteen packets for HTTPS, breaks down DNS lookup, TCP three‑way handshake, TLS 1.3… | ||
| Beyond the Basics - Advanced GIL Mitigation Patterns | Rohan Giriraj | 32m |
Shows concrete patterns: convert a NumPy RGB image to grayscale using Cython with nogil and OpenMP prange, compiled via cythonize into a .so and called from… | ||
| Automating License & Copyright Scanning in CI Pipelines With FOSSology | Rajul Jha | 27m |
Integrates FOSSology into CI pipelines (GitHub Actions, GitLab CI, Travis CI) via a published Docker image with prebuilt agents that runs Nomos, Ojo and… | ||
| Algorand Python: Next-Gen Smart Contracts Made Simple | Astha Yadav | 31m |
Claims a Algorand Python framework that is statically typed, compiles Python → PUYA AST → AWST → TEAL, and runs on Algorand with limits (no exceptions, no… | ||
| Scaling Python for Performance: 40x Faster Image Processing with Smart designing | Shaleen Jain | 30m |
Greedy rectangle-partitioning algorithm converts segmentation maps into exact non-overlapping rectangles and replaces deep recursion with a BFS task queue that… | ||
| Green AI at Scale: Energy-Efficient LLM Serving using vLLM & LLM Compressor | Abhijit, Anindita | 30m |
Shows concrete techniques and measured gains: vLLM uses PagedAttention plus speculative decoding and continuous batching to cut KV cache fragmentation and… | ||
| Surprises, Pitfalls, and Patterns: Learnings from Interviewing 400+ Developers | Vivek Keshore | 30m |
Over 400 interviews produce concrete failures not vague topics: 29% selection rate, 25% could write a decorator yet only 10% handle decorator-with-args and 5%… | ||
| DuckDB is the New Pandas | Anand S | 28m |
Claims that DuckDB replaces Pandas with concrete evidence: a 17M-row groupby distinct on parcels runs 0.2s vs 4s (≈20–25× faster), uses all cores, and reads… | ||
| MCP Unlocked: The Key to Smarter AI Contextualization | Sasidhar Donaparthi | 30m |
An open protocol that standardizes LLM access to external tools and read‑only data, MCP uses JSON-RPC 2.0 over transports like streamable HTTP and demoed a… | ||
| Building AI Apps That Don't Break: Architecture Lessons from HappyFox | Barath C, Anand KS | 31m |
HappyFox describes concrete engineering choices: a production AI suite (chatbot, copilot chat, writing assistant, ticket summary, KB draft) used by ~12,000… | ||
| [Panel] Vibe Coding: Yay or Nay? | Anand S, Anindita, Anirudha, Anand C, Shreya | 34m |
Panelists debate concrete gains and risks: prototypes that once took weeks now appear in hours using Cursor, Claude Code and GPT-5, a walking-phone… | ||
| Breaking Down Data Silos: Enterprise Data Automation with Python on Azure | Aakash Khandelwal | 32m |
Built a central Azure Data Lake repository that removes direct source access and integrates 20+ systems so Python jobs (pandas, Polars, DuckDB) handle… | ||
| Memory is the Agent: Architecting Stateful Reasoning | Archit Singh | 24m |
Shows a concrete memory layer (Python + Postgres demo on GitHub) that flips a stateless LLM into a stateful agent by read/write memory per turn; A/B demo… | ||
| Evals First, Code Later: A Practical Guide to Evaluations, Rerankers & Caches | Saksham Aggarwal | 30m |
A production code-search engine built with Python, PostgreSQL (pgvector) and OpenAI rerankers reduces average query latency from 20 minutes to 30 seconds — a… | ||
| Day 2 Lightning Talks | Subhajit Mukherjee | 44m |
A 5‑minute lightning talk that describes practical observability for agentic AI: a financial agent system is instrumented to surface failures across… | ||
| Day 2 Closing Address | Nabarun Pal | 13m |
Conference reports 1,694 attendees so far, with 1,270 conference attendees, 360 workshop seats and 70 Dev Sprint tickets left; Dev Sprint runs at Scaler School… | ||
| Edge ML with MicroPython + TinyML: Real-time Inference on under ₹500 Boards | Pratik, Sneha | 27m |
Runs a fall-detector pipeline that collects MPU6050 accelerometer samples (50-sample windows), extracts magnitude mean/std and axis traces, trains a tiny… | ||
| Learning and Coding in the Age of AI | Reuven Lerner | 43m |
Argues that modern AI is "the new electricity," helps but doesn't replace programmers, and hallucinates (it invents non‑existent methods); models know more… | ||
| Day 1 Lightning Talks | Akash Sil | 38m |
A personal project called VGIT is a terminal-based Git visualizer that installs via a GitHub clone and pip install -e (not yet on PyPI) and is currently a… | ||
| Day 1 Opening Address | Nabarun Pal | 09m |
Conference opens with practical logistics: 40 talks, 13 workshops, 11 poster sessions, 10 community booths, and a Dev Sprint on 15th at Scaler College… | ||
| $ python build reactor.py | Shakthi Kannan | 30m |
Shows that Python actually runs reactor-scale simulations: OpenMC (Monte Carlo with C++ kernels and MPI/OpenMP), Paramak for 3D tokamak geometry, and DAC MC… | ||
| nothing matches | ||