Skip to content
Rohith Anumalasetty
Résumé

Software engineer · AI systems · Hyderabad

AI systems, builtall the way through.

I'm Rohith, a CS undergrad at VNRVJIET and an AI Engineer Intern at Centella AI Therapeutics. I work across the whole system: FastAPI services, agent workflows, ML models, and the interfaces people use them through.

See selected workGet in touch

Now

AI Engineer Intern, Centella AI Therapeutics

Async backend services and LLM workflows for production AI features. Since Aug 2026.

Hackathon podiums
4
Two wins, two runner-ups
Springer paper
1
ICACECS 2025
CGPA
8.95
B.Tech CSE, VNRVJIET
TG EAPCET rank
2,499
of ~250,000 candidates

Runner-up — VJ Hackathon 2K26 ; 1st place — Convergence 2K26 ; 1st place — Webathon 4.0 ; First runner-up — VNR Design-A-Thon ; Published — Springer, ICACECS 2025 ;

01Selected work

Work that placed, and how it works.

13 projects. The four hackathon builds come first, with the systems behind them.

01

1st place · Convergence 2K26

Recoverix

In-transit shipment recovery

Problem
When a shipment breaks down in transit, operators need a viable recovery option quickly, and they need to trust why that option was chosen.
Approach
The logistics network is modelled as a graph in NetworkX. Candidate recoveries are filtered against hard constraints, every decision carries an explanation, and the chosen route is handed off through Sarvam voice automation with Google Maps geospatial routing.
core capabilities in one pipeline
5
hackathon build window
24h
  • Python
  • NetworkX
  • Google Maps
  • Sarvam AI
  • Explainable AI
Network graphfig. 01 · schematic
ORIGINDESTDISRUPTEDRECOVERY ROUTE
plannedre-plannedconstraints ✓ · explained ✓ · voice handoff ✓
02

1st place · Webathon 4.0

Civix

Civic micro-task platform

Problem
A broken streetlight or a faded notice rarely gets reported, because reporting feels like a commitment nobody signed up for.
Approach
Issues are broken into quick tasks. AI validates each submission and automation workflows move the issue forward, so contributing takes minutes instead of follow-ups.
civic issues tracked
30+
users reached
50+
  • Next.js
  • MongoDB
  • AI agents
  • Automation workflows
micro-task-three.vercel.app
Civix live site, home screen
03

First runner-up · VNR Design-A-Thon 2026

Traffix

AI traffic optimisation and prediction

Congestion forecast · 16 areasfig. 03 · schematic
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
N–Sphase
E–Wphase
Turnphase
Pedphase

YOLOv8 → scikit-learn → SUMO

Problem
Fixed signal timings cannot react to how traffic actually moves through a city like Bengaluru.
Approach
YOLOv8 and OpenCV detect vehicles, scikit-learn models predict congestion, and SUMO simulation tests the adaptive signal-timing recommendations before they are served through FastAPI and PostgreSQL.
  • YOLOv8
  • OpenCV
  • scikit-learn
  • FastAPI
  • PostgreSQL
  • SUMO
rows of live traffic data
250k
city areas analysed
16+
registrations in the field
2,800+
04

Runner-up · VJ Hackathon 2K26

PondFix

Early warning for aquaculture ponds

Problem
A pond can lose its oxygen overnight, and by the time a farmer sees fish gasping at the surface the damage is done. Readings alone do not say what is about to happen.
Approach
Every reading is validated, classified against one set of risk rules and settled into a persistent alert lifecycle, then a hash-pinned scikit-learn model forecasts +15, +60 and +180 minutes so oxygen alerts can open before the drop. An AI advisor and a RAG chatbot explain alerts in Telugu and twelve other languages, with Sarvam voice calls for critical ones, on a Next.js dashboard with a 3D digital twin and an Expo app for farmers.
ponds in the digital twin
17
historical observations replayed
72,750
forecast horizon
+3h
hackathon build window
24h
  • Next.js
  • Node.js
  • MongoDB
  • scikit-learn
  • Three.js
  • Expo
  • Chroma
  • Sarvam AI
Dissolved O₂ · pond 7fig. 07 · schematic
+15m+60m+180mWARNINGCRITICALNOWFORECAST · NOT OBSERVED
observedforecastpredicted alert → advisor → voice call

More systems

Retrieve → rank → reportfig. 04 · schematic
  1. 01
  2. 02
  3. 03
  4. 04
  5. 05
  6. 06
  7. 07
  8. 08
  9. 09
  10. 10

PatentPilot

AI systems

Retrieves relevant patents from public molecular databases, ranks the top five with a custom scoring pipeline, and writes structured patentability reports with overlap and novelty-risk assessments.

10 patents retrievedTop 5 ranked and reported

  • Next.js
  • FastAPI
  • MongoDB
  • LangChain
  • Gemini
Multi-train journeyfig. 05 · schematic
TRAIN ATRAIN BTRANSFERDELAY RISK

RailFix

AI systems

Recommends multi-train journeys with delay-risk analysis, intermediate routing, accessibility filters and interactive maps.

8,000+ stations13,000+ trains

  • Next.js
  • Python
  • MongoDB
  • n8n
  • Gemini

Private repository

K-Means + KNNfig. 06 · schematic

EF-BER

ML research

Published · Springer, ICACECS 2025

A framework combining K-Means and KNN for large-scale EDTA-based water-hardness classification that outperformed Random Forest and SVM benchmarks.

0.0020 mean error rate70,298 samples in the largest dataset

  • Python
  • K-Means
  • KNN
  • Research

Smaller builds

Full project index

02Profile

I like problems that need the whole stack.

Most of what I build sits where AI meets a hard real-world constraint: a shipment that still has to arrive, a traffic signal that has to change, a loan decision someone deserves to have explained.

That means caring about the unglamorous parts as much as the model. Constraint filtering before recommendations, explanations next to predictions, and backend services that make an agent workflow usable by an actual product.

Hackathons are where I learn fastest: two first places and two runner-up finishes so far, each one a full system built under a deadline. Alongside that I co-authored a Springer-published paper on water-quality classification, and I'm now building async FastAPI services and LLM workflows at Centella AI Therapeutics.

Studying
B.Tech CSE at VNRVJIET, Hyderabad. 2024 to 2028, CGPA 8.95.
Working
AI Engineer Intern, Centella AI Therapeutics. Remote.
Published
Springer, ICACECS 2025 proceedings. Co-author.
Giving back
ISTE volunteer. Guided 30+ juniors through a web development workshop.
Speaks
English, Telugu, Hindi.

Shipping cadence · last 12 months

github.com/nxtrohith751 contributions in the last year

03Stack

The tools I reach for, grouped by the job they do.

Languages

What I think in

  • Python
  • TypeScript
  • JavaScript
  • Java
  • C++
  • C

Product & interface

Where users meet the system

  • Next.js
  • React
  • Tailwind CSS
  • HTML / CSS
  • Bootstrap

Backend & data

Services, APIs, storage

  • FastAPI
  • Node.js
  • Express
  • REST APIs
  • MongoDB
  • PostgreSQL
  • Supabase

AI & automation

Agents, models, workflows

  • LangChain
  • LangGraph
  • AI agents
  • Prompt engineering
  • n8n
  • scikit-learn
  • Gemini
  • Llama

Tooling & cloud

Cloud at a basic level

  • Git
  • Docker
  • PostHog
  • Vercel
  • Azure
  • AWS
  • Google Cloud
  • Claude Code

04Experience

Where I've been building, and what it earned.

  1. Aug 2026 — Present

    AI Engineer Intern

    Centella AI Therapeutics · Remote

    Building asynchronous backend services and REST APIs with FastAPI and asyncio, and integrating LLM-driven workflows into production-facing AI product features.

    • FastAPI
    • asyncio
    • LLM workflows

  2. 2025

    Agentic AI, semester-long program

    P. Sai Teja · VNRVJIET · Hyderabad

    A semester on agentic systems and automation: MCP, handoffs, guardrails, evals and AI benchmarks, with hands-on work in LangChain and LangGraph.

    • LangChain
    • LangGraph
    • MCP
    • Evals

  3. 2024 — 2028

    B.Tech, Computer Science & Engineering

    VNR Vignana Jyothi Institute of Engineering and Technology · Hyderabad

    CGPA 8.95. Volunteer with ISTE, where I guided 30+ juniors through a web development workshop.

Recognition

05Contact

Local time in Hyderabad:

Got a hard problem? I'd like to hear it.

[email protected]