Open to software engineering opportunities

Software Engineering & AI @ IIT Kharagpur

I build reliable systems that perform under pressure

Backend systems, algorithms, and applied AI - engineered around reliability, latency, throughput, and real-world usefulness.

Backend SystemsAlgorithmsApplied AI
View projects View résumé GitHub
Portrait of Ayush Goel
Ayush GoelFinal-year undergraduate
system.profileONLINE

$ build → benchmark → improve

Scroll to inspect selected workIND / 2026

01 / IMPACT

Measured, not just shipped

Selected outcomes from systems, document intelligence, and algorithmic work.

164 RPSLLM gateway throughput

1,600-request benchmark with zero failures.

 86.7%Lower p50 latency

Redis-cached requests: 81.2 ms → 10.8 ms.

 75×Faster repeat OCR

SHA-256 extraction cache: 30.1 s → 0.4 s.

 500+Algorithmic problems

Codeforces Expert with a 1638 peak rating.

02 / ABOUT

Software first
AI where it earns its place

I’m a final-year undergraduate at IIT Kharagpur, pursuing a B.Tech. (Hons.) in Chemical Engineering with a Micro-specialization in Artificial Intelligence and Applications.

My core interests are software engineering, backend systems, algorithms, and system design. I’ve applied that foundation across machine learning, computer vision, reinforcement learning, document intelligence, climate forecasting, and robotics.

01Design for failureRetries, fallback, observability
02Measure the systemLatency, throughput, correctness
03Automate confidenceTests, CI, reproducible benchmarks

03 / SELECTED WORK

Systems, intelligence,
and algorithmic depth

Three projects that best represent how I think about architecture, performance, and reliability.

MODELROUTE / LIVE TRACEHEALTHY
cache.hit10.8ms200 OK
01 / BACKEND SYSTEMSFLAGSHIP PROJECT

ModelRoute

Resilient Multi-Provider LLM API Gateway

A production-oriented gateway that keeps model traffic moving when providers slow down, fail, or hit limits.

  • Three routing strategies across OpenAI and Gemini adapters with bounded retries and automatic fallback.
  • Async FastAPI handling, Redis caching, atomic Lua rate limiting, and PostgreSQL p50/p95 observability.
81.2 → 10.8 mscached p50164 RPSpeak throughput0 failures1,600 requests
FastAPIRedisPostgreSQLDockerAsync Python
Explore repository
PROCTORVISION / REVIEW3 EVENTS
gaze_away 02.4s multiple_faces cooldown phone_detected review
02 / SOFTWARE + VISIONREVIEW PLATFORM

ProctorVision

Vision-Guided Proctoring Review Platform

A review-first computer-vision system that turns noisy frame detections into persistent, inspectable behavioral events.

  • AlertEngine with five duration-aware rules, grace periods, and cooldown controls.
  • Primary-candidate tracking combines IoU, center distance, and area similarity; validated through CI.
~19%higher throughput279.4 → 234.9 msframe latency79 testsautomated validation
FastAPIMediaPipeYOLOOpenCVGitHub Actions
Explore repository
PRM / SEED 42PATH FOUND
nodes 502edges 5,312search 1.5 ms
03 / ALGORITHMS + ROBOTICSDETERMINISTIC BENCHMARKS

Deterministic PRM + A*

Sampling-Based Motion Planning & Graph Search

A reproducible planning stack that constructs collision-safe roadmaps and measures graph-search behavior against a goal-biased RRT baseline.

  • PRM roadmap generation, KDTree neighbor search, conservative collision checking, and Euclidean A*.
  • Matched-seed PRM vs RRT evaluation across 60 planning attempts.
502 nodes500 samples + anchors1.5–19.6 msA* search73 testsalgorithm + pipeline
PythonSciPyNumPyKDTreeA*OpenCV
Explore repository

04 / EXPERIENCE

Engineering across product and research

Focused stories, not a résumé transcript: what I built, the constraints I worked with, and what changed.

Open full résumé
01

Wasserstoff RJ Innovations Pvt. Ltd.

Software Development Engineer Intern

Built a Python/FastAPI document-intelligence pipeline with Tesseract, Docling, OpenCV preprocessing, Docker, concurrent batch execution, and SHA-256 extraction caching.

83%CER reduction0.481 → 0.080character error rate75× fasterrepeat extraction
02

University of Manchester × University of Liverpool

Undergraduate Research Intern

Engineered Xarray/ERA5/ChaosBench pipelines for subseasonal Z500 and T850 forecasting, adapting ArchesWeather-S with a compact residual CNN head.

225daily Zarr stores0.23%parameters trained29M+predictions evaluated
03

Chi SquareX Technologies

Machine Learning Developer Intern

Designed an object-oriented options simulator in OpenAI Gymnasium and trained PPO/A2C policies with Stable-Baselines3 on ten months of ADANIPOWER options data.

54.05%historical backtest return48%win rate16%maximum drawdown
Historical backtest results; not indicative of future performance.
+

Aerial Robotics Research Group, IIT Kharagpur

Undergraduate Researcher

Worked across ROS, Gazebo, MAVROS, Iris drone simulation, keyboard-controlled 3D navigation, edge detection, and the Hough Line Transform.

05 / ALGORITHMIC SIGNAL

CODEFORCES
1638Peak ratingEXPERT

500+ problems
One habit: reason precisely

Problem-solving across Codeforces, LeetCode, AlgoZenith, and AtCoder, with notable global ranks of 1,863, 2,320, and 3,704 in rounds with 27K+ contestants.

View ayushgoel02
SELECTED GLOBAL RANKSLOWER IS BETTER
#1,86327K+ field
#2,32027K+ field
#3,70427K+ field
C++DSACONTESTS500+

06 / TOOLKIT

Backend-weighted
Problem-driven

I choose tools for reliability and fit, with strongest emphasis on software systems and backend engineering.

CORE / BACKEND01

Systems I reach for first

PythonC++FastAPIPostgreSQLRedisREST APIsNode.jsExpress.jsMongoDBSQLJavaScriptC
ENGINEERING / DEVOPS02

Shipping with confidence

LinuxDocker ComposeGitHub ActionspytestGitGitHubPostman
APPLIED AI03

Intelligence as a capability

PyTorchOpenCVNumPySciPyMediaPipeYOLOHugging FaceXarrayStable-Baselines3

07 / RECOGNITION

Small section
Strong signal

#07

Convolve 4.0 · Pan-IIT AI/ML Hackathon

All-India Rank 7 in Round 1

National inter-IIT machine-learning competition.
2026
#22

Optiver Quantitative Trade-a-thon 2026

Real-time market making and pricing

Designed strategies under dynamic simulated market conditions.
2026

CURRENT FOCUS / NOW

Building depth in scalable backend systems, distributed-system fundamentals, and production-grade software engineering

System DesignAPIsAlgorithmsOOP / LLDOSDBMSNetworkingApplied AI

08 / CONTACT

Let’s build something
that holds up

I’m interested in software engineering, backend, and applied AI opportunities where thoughtful systems work matters.

Start a conversationgoel85360@gmail.com
LinkedIn ↗GitHub ↗Codeforces ↗