Aarif Khan

Resume

Software Engineer, Tekion Corp. — Bengaluru, India — Jan 2025–Present (joined as Software Engineering Intern)

Designed and delivered modular, plug-and-play webhook infrastructure processing 500 Kafka records/sec, enabling new publishers to onboard by adding a single class — no changes to existing code
Reduced p99 latency from 400ms to under 200ms (a 50% improvement) by root-causing and resolving a p99 latency SLA breach on a production open API GET endpoint through query optimization.
Decreased average latency from 350ms to 10ms (a 97% improvement) on a read-heavy config API handling 50 req/sec by caching infrequently-changed config values in memory.
Owned backend design and implementation for a GM vehicle check-in feature end to end, building a screen-agnostic job-creation API that consolidated 4 separate screen integrations into a single reusable contract with no per-screen logic
Built a Systems Integrator platform enabling third-party companies to develop integrations on Tekion’s open APIs — deployed in an isolated cluster for security, reducing integration go-to-market time by shifting delivery from a single internal team to multiple third-party teams working in parallel, validated by delivering the first proof-of-concept integration with a small team
Diagnosed and fixed a P0 production incident where a warranty posting defect caused ~2,000 invalid entries for the first dealer migrated to a new system (out of 2,000 total customers);

Associate Software Developer, Location Solutions, LLC — Dubai (Remote, Part-time) — Mar 2024–Aug 2024

Rewrote a legacy .NET/C# service in Node.js, improving performance, scalability, and maintainability; reduced average response time by 20% for a specific workload, used by every Location Solutions customer
Implemented in-memory join algorithms for time-series application data, avoiding repeated single-record database read/write round-trips of the database.
Increased backend algorithm test coverage to 60% through comprehensive unit testing, improving reliability and reducing regressions
Integrated OpenTelemetry logging into a legacy service, enabling retroactive debugging of customer-reported failures that previously had no visibility.

Schema-Guaranteed LLM Decoding for Structured Medical Extraction — Independent project — Source

Tools & technologies used: Python, Transformers, XGrammar, Pydantic, Qwen2.5-1.5B.
Built a grammar-constrained LLM pipeline using XGrammar to compile a JSON schema into GPU-level token constraints, guaranteeing schema-valid structured output (symptom, severity, duration, category) from unstructured clinical notes.
Validated 200 real clinical notes end-to-end, measuring structural validity (JSON parses and matches schema) separately from semantic accuracy (extracted category matches ground-truth medical specialty), then ran failure analysis on the misses to trace them to model capacity rather than the schema constraint.

Real Time Object Detection on a Video Stream from Drone/Mobile — Team/hackathon project, built during CodeUtsava 7.0 — Source

Problem solved: pothole detection using drone footage and machine learning.
Tools & technologies used: MERN stack, Flask, OpenCV, YOLOv8, RTMP protocol.
Took an input stream from an RTMP source and ran object detection on the video frames.
Built the detection script with OpenCV to run object detection and send frames to the server and database.
Built an API to store detected frames into the file system and MongoDB database.
Built a frontend UI in React to interact with the detected objects.

CNN-Based Haze Classification and Dehazing — Independent project, Bachelor’s minor project — Source

Tools & technologies used: Python, TensorFlow/Keras, CNNs, Res2Net, PixelShuffle.
Built a 3-class CNN classifier (No Haze / Mild / Dense) to route hazy images to a haze-level-appropriate dehazing network, reaching 93% validation accuracy, up from 37% at epoch one.
Implemented a teacher-student knowledge-transfer dehazing network (based on Wu et al.’s KTDN architecture) from a research paper with unspecified training details, combining L2, SSIM, and VGG19 perceptual losses for the teacher network.
Trained the teacher network on a combined RESIDE/I-Haze/O-Haze/NH-Haze dataset, reducing validation loss from 0.40 to 0.15 over 200 epochs.

B.Tech in Computer Science and Engineering — National Institute of Technology, Raipur — 2025 CGPA/Percentage: 7.52/10

Languages: C++, JavaScript, Java, Python

Developer Tools: Git, Postman, Chrome Developer Tools

Frameworks: Spring Boot, Express.js, Node.js

Machine Learning: TensorFlow/Keras, Hugging Face Transformers, CNNs

Cloud / Databases: SQL, MongoDB, Kafka, Redis, Elasticsearch

Soft Skills: Communication, Team Collaboration, Problem-solving, Analytical Thinking

Areas of Interest: Operating Systems, Databases, Computer Networks

Winner, CodeUtsava 7.0 — Central India’s largest hackathon, organized by NIT Raipur — Nov 2023
Winner, CSE-MCA-IT branch, Vigyan Science Exhibition — organized by NIT Raipur — Oct 2023
Codeforces Specialist — Global Rank 530 in Codeforces Round 944 (Div. 4) — May 2024
Global Rank 2446/29648 in LeetCode Biweekly Contest 112