gdg chennai hackfest '26 · feb 2026
build with ai:
what, why & how
ideas, apis, and everything you need to ship something incredible today
sanju
founder, thisux · droidclaw · unitedby.ai
gdg chennai hackfest '26 · feb 2026
ideas, apis, and everything you need to ship something incredible today
sanju
founder, thisux · droidclaw · unitedby.ai
what's this talk about
this is a practical walkthrough — from what ai actually is, to what you can build with it today.
by the end, you'll have real project ideas and the tools to build them.
who's talking
thisux
design studio
dunsocial
ai product
dun
ai-native tools
droidclaw
went super viral
unitedby.ai
ai community
i build with ai every day. agents, automations, products. let me show you how.
and who's built something with an api? keep your hand up.
let's start from scratch. no jargon.
the simple version
ai sees billions of examples and learns to predict the next thing.
text? it predicts the next word.
images? it predicts the next pixel.
code? it predicts the next line.
that's it. really.
the landscape
narrow ai (ANI)
one task, really well
chatgpt, gemini, dall-e. this is where we are today.
general ai (AGI)
human-level thinking
can do any intellectual task a human can. not here yet.
super ai (ASI)
beyond human
smarter than all humans combined. science fiction... for now.
the agi question
short answer: no one really knows.
2027-2030
optimistic estimates (openai, google deepmind)
2040+
conservative estimates (most researchers)
but what we have today? already incredibly powerful for building things.
the model landscape
gemini
multimodal, free tier
gpt-4o
openai
reasoning, tools
claude
anthropic
coding, analysis
llama 3
meta
open source
mistral
mistral ai
open source, fast
deepseek
deepseek
reasoning, cheap
grok
xai
real-time data
gemma
lightweight, open
the big picture
people using ai tools globally
and that number grows by millions every month. your users are already there.
from simple scripts to fully autonomous agents
let's define it
it's not just "make a chatbot." it's removing human effort from repetitive, predictable tasks.
real automation =
trigger → decision → action → feedback loop
email comes in → ai reads it → drafts response → you approve (or don't)
the spectrum
simple rules
if-then automation. zapier, ifttt. no ai needed.
ai-assisted
ai suggests, human decides. copilot, grammarly, smart compose.
human-in-the-loop
ai does the work, human approves. best for most hackathon projects.
supervised autonomy
ai acts independently, human monitors. ai agents with guardrails.
full autonomy
ai handles everything end-to-end. self-driving cars. very few systems here.
the sweet spot
ai does the heavy lifting
drafts, analyzes, summarizes, generates code
human validates
reviews, approves, rejects, or edits the output
system improves
feedback loop makes the ai better over time
this is where 90% of useful ai products live. and it's perfect for a hackathon.
why it matters
27%
of ai-generated code has bugs (github study)
15-20%
of factual claims by llms are wrong
human oversight isn't optional. it's the product feature.
the frontier
agents that can plan, use tools, and execute multi-step tasks on their own.
what they can do
browse web, write code, call apis, manage files, send emails, book meetings
what they struggle with
long-term planning, ambiguous goals, error recovery, knowing when to ask for help
think of agents as interns: capable but need clear instructions and supervision.
how agents work
// simplified agent architecture
while (task_not_complete) {'{'}
observe() // read context
think() // plan next step
act() // call tool or api
evaluate() // did it work?
{'}'}
this is the mental model. every agent framework follows this pattern.
where ai is making the biggest impact right now
sector 1
diagnosis support
ai that reads medical images, flags anomalies, assists doctors
patient engagement
appointment scheduling, symptom checkers, follow-up reminders
drug discovery
protein folding, molecule screening, clinical trial matching
hackathon idea
ai-powered medicine reminder + interaction checker using gemini
sector 2
personalized tutoring
adapts to each student's learning pace and style
content generation
quizzes, flashcards, lesson plans from any source material
accessibility
translating content, sign language, text-to-speech, simplification
hackathon idea
ai study buddy that reads your notes (drive api) and quizzes you (gemini)
sector 3
email automation
smart replies, categorization, priority inbox on steroids
document processing
summarize contracts, extract data, auto-fill forms
meeting assistant
transcribe, summarize, extract action items, schedule follow-ups
hackathon idea
gmail + gemini agent that auto-drafts responses based on your style
sector 4
route optimization
delivery routes, traffic-aware scheduling using maps api
local discovery
ai-powered "what's near me" with reviews and recommendations
inventory management
predict demand, auto-reorder, waste reduction
hackathon idea
ai trip planner: maps api + gemini for personalized chennai itineraries
and there's more
agriculture
crop monitoring, soil analysis
fintech
fraud detection, budgeting
entertainment
content recs, music gen
real estate
price prediction, virtual tours
cybersecurity
threat detection, phishing
e-commerce
product recs, support bots
legal
contract review, research
transport
routing, demand prediction
free gcp credits. powerful apis. let's build.
the good news
you have google cloud credits. use them. build something real.
gemini api
gmail api
maps api
drive api
api deep dive
the brain of your project
what it does
text generation, code writing, image understanding, audio processing, function calling
why it's great for hackathons
generous free tier, multimodal out of the box, google ecosystem integration
// get started in 3 lines
import google.generativeai as genai
model = genai.GenerativeModel('gemini-pro')
response = model.generate_content("hello!")
api deep dive
automate the inbox chaos
read emails
fetch, search, filter, label — programmatically
send emails
compose, reply, forward — with ai-generated content
organize
auto-label, categorize, archive — based on ai analysis
combine with gemini: ai reads your email, understands context, drafts the perfect reply.
api deep dive
location intelligence for your app
places
search nearby, get details, photos, reviews
directions
routes, distance, traffic, eta, waypoints
geocoding
address ↔ coordinates, reverse geocoding
pair with gemini for natural language: "find me the best biryani within 2km that's open now"
api deep dive
your ai's file system
read files
access docs, sheets, pdfs — feed them to gemini
create & update
generate reports, update spreadsheets, create docs
organize
auto-organize files, smart search, permissions
imagine: upload a pdf → ai extracts insights → creates a summary doc → emails it to your team
putting it all together
combine apis into something greater than the sum of its parts
gemini as the brain — understands intent, generates responses
gmail as the communication layer — reads and sends emails
drive as the knowledge base — stores and retrieves documents
maps as the location layer — adds spatial awareness
the secret sauce
how ai talks to your apis
// tell gemini what tools it has
tools = [
search_emails(query),
send_email(to, subject, body),
find_nearby_places(location, type),
read_drive_file(file_id)
]
// gemini decides which tool to use
user: "email the team about tomorrow's meeting venue"
gemini → calls find_nearby_places() → then send_email()
the model decides which api to call. you just define the tools.
the architecture
steal these. seriously. that's what they're here for.
what it does
reads your gmail inbox, categorizes emails by urgency, drafts contextual replies, learns your tone
apis needed
gmail api (read/send) + gemini api (analysis/generation)
human-in-the-loop: ai drafts → you review → one click to send or edit
what it does
upload notes to drive, ai reads them, creates flashcards, generates practice questions, tracks progress
apis needed
drive api (file access) + gemini api (content generation + q&a)
bonus: add gmail api to email daily study reminders with quiz questions
what it does
natural language trip planner. "plan me a weekend in chennai with street food and temples"
apis needed
maps api (places + directions) + gemini api (planning + conversation)
bonus: add drive api to save itineraries, gmail to share with friends
what it does
connects to google drive, reads all your docs, lets you ask questions across all files at once
apis needed
drive api (file listing + reading) + gemini api (summarization + q&a)
example: "what did the marketing team promise in last month's reports?"
what it does
camera input → gemini describes the scene → text-to-speech output. helps visually impaired navigate
apis needed
gemini api (vision + text) + maps api (navigation) + text-to-speech
impact: this kind of project wins hackathons because it solves a real human problem
what it does
one agent that handles: email triage, file organization, meeting prep, daily briefings
apis needed
all of them! gmail + drive + maps + gemini. the full google ai stack.
scope tip: start with one flow (email triage). add more if time permits.
ask yourself:
1. what annoys me every day?
2. what task do i keep postponing?
3. what would help someone i know?
4. what data do i already have access to?
the best hackathon projects solve real problems. yours.
pro tips
start small, demo big
nail one feature perfectly. don't build half of everything.
demo > code
judges care about what it does, not how clean your code is.
use ai to build your ai project
use gemini, claude, copilot to write code faster. that's the whole point.
tell a story
"we built x because y" always wins over "we used these 12 apis"
let's go
# 1. set up gcp project
gcloud projects create my-hackathon-project
# 2. enable apis
gcloud services enable generativelanguage.googleapis.com
gcloud services enable gmail.googleapis.com
gcloud services enable maps-backend.googleapis.com
gcloud services enable drive.googleapis.com
# 3. get api key
gcloud auth application-default login
# 4. install sdk
pip install google-generativeai google-api-python-client
resources
gemini api docs
ai.google.dev
gmail api quickstart
developers.google.com/gmail/api
maps platform
developers.google.com/maps
drive api quickstart
developers.google.com/drive/api
google cloud console
console.cloud.google.com
build something that makes one person's life easier.
not "revolutionary ai platform." not "gpt wrapper."
something specific. something useful. something you'd actually use tomorrow.
questions? ideas? come talk to me.
sanju
sanju.sh — personal
thisux.com — design studio
droidclaw.ai — went viral
dunsocial.com — ai product
unitedby.ai — ai community
gdg chennai hackfest '26
let's build with ai. together.