IIT Tirupati · B.Tech CSE · Class of 2028

Ankit
Sinha

I build backend systems and applied-AI tools.

I build streaming data pipelines, RAG, backends, and agentic systems — using AI where it helps and owning the hard decisions myself.

100,000+installs · ScraGet A published Python package I built as a teenager — real distribution, still installed today.
8.93CGPA · IIT Tirupati B.Tech Computer Science & Engineering, Class of 2028.

01 About

Shipping since before college.

I taught myself to program in the Scratch community as a kid. That turned into ScraGet — a Python package with 100,000+ installs — and community bots run at real scale, all before college.

Now I'm a CSE undergrad at IIT Tirupati, where I help lead the coding club and mentor juniors — while shipping streaming pipelines, RAG systems, and embedded firmware.

The through-line: I use AI where it helps and own the hard decisions myself. I'm aiming at backend and applied-AI roles where the system actually has to work.

02 Results

Four national placements.

Across systems, RAG, and applied-AI competitions.

2nd
of 1,815 teams
Pathway Hackathon, IIT Ropar
Real-time RAG research assistant
1st
place
Code for Society, IIT Tirupati
Legal-compliance RAG — AttorneysInRAGs
5th
of 23 IITs
Inter IIT Tech Meet 14.0
Vigil360 — real-time KYC pipeline
5th
of 400+ teams
Orion Astrathon, IIT Mandi
National finalist

03 Work

Scoped to exactly what I built.

Team projects below credit only my own contribution.

Tier 1

Flagships

The three projects that define the backend × applied-AI story.

Solo backend

AttorneysInRAGs

Legal-compliance analysis with RAG

Scores a Terms-of-Service document by how many rules from a legal rules database it actually satisfies.

Owned The full backend, end to end: a rules database with severity/type/source classification; a DistilBERT classifier that strips legal filler; stop-word and duplicate-removal preprocessing; hybrid keyword-and-similarity rule retrieval; a local quantized LLM (Llama 3 / Mistral via Ollama) doing the compliance scoring, with context-window packing for long documents; and Python risk scoring on top.
Stack PythonDistilBERTOllamaLlama 3 / Mistral
Result 1st place — Code for Society, IIT Tirupati

Solo · Published package

ScraGet

Python wrapper for the Scratch APIs

A published package that abstracts the Scratch web APIs into a clean Pythonic interface, cutting boilerplate by ~70%. Built as a teenager, installed 100,000+ times over its lifetime — real distribution, real users, real maintenance.

Owned Everything — API design, implementation, releases, and docs.
Stack PythonREST APIs
Result 100,000+ installs

Tier 2

Systems & ownership

Lower in the stack, or bigger in scope.

2-person course project

Pipelined RISC-V Processor

A 5-stage pipelined RISC-V processor with full hazard forwarding and a three-level cache hierarchy with LRU replacement.

Owned Designed the pipeline and the forwarding logic; defended the design in a live viva.

Solo · Real users

JoSAA Data Pipeline

A scraping and parsing pipeline that turned official JoSAA reports and PDFs into clean, queryable admissions data — 110,000+ rows processed, used by 6,000+ students during counselling, with 66,000+ impressions across student forums.

Owned All of it — solo build.
Stack PythonBeautifulSoupMySQL

Led 12-person team

Sensor-Enabled CPR Training Manikin

A low-cost CPR training manikin built for ~₹3,670 against ₹30,000+ commercial units — roughly 10× cheaper — with live feedback to a companion mobile app.

Owned Led the 12-person team end to end (hiring and team formation, architecture, budget and parts, dispute resolution, integration) and personally built the ESP32 firmware: sensor acquisition, calibration, and BLE telemetry, in Embedded C++. The companion app was a teammate's build.
Stack ESP32Embedded C++BLE

Tier 3

Ideas & teaching

Thinking in public: AI systems design, research, and writing.

AI coaching system

Pointer Invariant Coach

An AI coaching system — a folder of markdown, deliberately — that walks competitive programmers through five gates and won't let them write a line of code until they can state their loop invariant precisely.

Result Honorable mention — Jake's Weekly Comp

Research · Conceptual

The Observer Paradox

Adaptive software

A research project on reducing software bloat without runtime monitoring — the core argument being that the monitoring often costs more than the bloat it manages. Delivered as a markdown knowledgebase and a multi-agent orchestration design; earned an S grade for the course.

Teaching · Writing

Technical Blog

State-space modeling & optimization

Long-form teaching posts for first-years. The flagship arc models tic-tac-toe's state space and shrinks it from 19,683 boards to 304 by exploiting symmetry — measured one optimization at a time.

Also maintained

  • saaranshbot — a paid Discord automation bot, built for a client and kept running in production.
  • SolarTracker — a sun-tracking rig: hand-written firmware, serial plotting, and 3D-printed parts I CAD-modeled myself.

04 Skills

The toolbox.

Languages

  • Python
  • C / C++
  • SQL

Backend & systems

  • FastAPI · Flask
  • Kafka · Debezium
  • Pathway
  • Docker
  • PostgreSQL
  • Linux

Applied AI

  • RAG
  • LLM inference
  • Quantization · Ollama
  • ChromaDB
  • Multi-agent orchestration

Tools

  • Git
  • GCP
  • NumPy
  • BeautifulSoup

05 Leadership & education

On campus.

  • Deputy Head — Digital WizardsCoding Club, IIT Tirupati
  • Deputy Head — GDG on CampusIIT Tirupati
  • Core Member — Sigma SquadAI/ML Club, IIT Tirupati
  • Contingent MemberInter IIT Tech Meet 14.0

Education

B.Tech, Computer Science & Engineering

IIT Tirupati · 2024–2028

CGPA 8.93 / 10

06 Contact

Let's talk.

Looking to be hired for a Summer 2027 internship — backend, platform, or applied-AI / agentic roles. The fastest way to reach me is email.