Skip to content
CK

Software Engineer / Student

hi, chetan here.

about

CK

Software Engineer

Here is a little background

I am a software engineer and computer engineering student who likes building systems that are usable, inspectable, and grounded in real technical constraints. My work sits across machine learning pipelines, embedded sensing, and product-facing applications, with a consistent focus on reliability and implementation depth.

experience

C-TEL InfoSystems Pvt Ltd logo

Software Engineer Intern @ C-TEL InfoSystems Pvt Ltd

May 2026 - Aug 2026 · Hyderabad, India

  • Shipped a Django and React internal ERP portal serving role-scoped views to staff, managers, and external vendors, verified by 346 backend and 1,743 frontend tests.
  • Enforced role-based access control across three visibility tiers, closing a cross-supplier document exposure found in adversarial review by fixing scoped file serving test-first.
  • Traced an 80-second production report to an unindexed join between a 4.3M-row and an 88M-row Oracle table, 39 seconds per depot on that join alone, where one index takes it to an estimated 2 to 3 seconds and unblocks 3 further reports that would not finish at all.
  • Removed 2 seconds per request from a vision service by profiling the request path down to name resolution, where the client tried IPv6 first against a v4-only host before falling back.
  • Built screens for the companion Flutter app on iOS and Android across leave, approvals, and procurement, consolidating its colours and text styles into one shared source so screens stop drifting visually.
DjangoReactOracleFlutterApache Superset

education

New York University logo

New York University

Master of Science in Computer Engineering

New York, USA | Sep 2025 - May 2027

Machine LearningDeep LearningMLOpsEmbedded Systems
King's College London logo

King's College London

Bachelor of Science in Computer Science (Artificial Intelligence)

London, UK | Sep 2021 - Aug 2024

Data StructuresOperating SystemsDatabasesCompilersMachine Learning

skills

Hover over a skill for current proficiency

Python logo

90%

C++ logo

80%

TypeScript logo

82%

React logo

84%

Next.js logo

76%

FastAPI logo

78%

Django logo

74%

Docker logo

75%

Kubernetes logo

66%

PyTorch logo

78%

MLflow logo

72%

ONNX logo

76%

PlatformIO logo

68%

BLE logo

66%

PostgreSQL logo

78%

Linux logo

78%

Scala logo

70%

C logo

74%

GitHub Actions logo

74%

Tailwind CSS logo

80%

Machine Learning

Deep LearningNatural Language ProcessingComputer VisionLarge Language Models

Systems and practices

REST API designMicroservicesRole-based access controlTest-driven developmentProfiling and index tuningWCAG 2.1 accessibility

projects

View GitHub →

American Option Lattice Engine

C++17CRR LatticeQuant Finance

C++17 binomial lattice pricer: Cox-Ross-Rubinstein for European options, Snell-envelope backward induction for American early exercise. Swapping the naive O(N^2) grid for an O(N) rolling path took a 50,000-step American put from 173 seconds and 9.7 GB to 1.6 seconds and 0.4 MB, with both paths agreeing to 1e-10. Validated against Black-Scholes convergence and put-call parity rather than against itself, with 30 checks gating CI. Testing surfaced 3 real defects, including an incomplete no-arbitrage check that was silently admitting a negative risk-neutral probability.

Source →

GLA-DPO Transformer

PyTorchGated Linear AttentionDPO

A 158M-parameter Gated Linear Attention language model written from scratch in PyTorch, pretrained on TinyStories and then aligned with Direct Preference Optimization, no separate reward model. GLA matches plain linear attention's loss at equal steps while staying O(N), and holds memory constant during inference where a standard transformer's KV-cache keeps growing. A two-person NYU project with Bryce Miranda: he built the DPO stage and preference data, I built the GLA architecture, pretraining, and the generation interface.

Source →

Two Compilers: LLVM-IR and JVM Bytecode

ScalaLLVM-IRJVM Bytecode

Two Scala compilers sharing a regex-derivative lexer built from first principles rather than a generator. One takes a small functional language to LLVM-IR end to end, demonstrated on recursion, Mandelbrot, and Towers of Hanoi. The companion repo, while-language-jvm-compiler, interprets a WHILE-like imperative language and emits Jasmin-style JVM assembly, demonstrated on Collatz, Fibonacci, factorization, and primes.

Source →

Parkinson's Motion Monitor

STM32 / mbedCMSIS-DSP FFTBLE GATT

About 1,445 lines of embedded C++ on an STM32 DISCO-L475VG-IOT01A, reading an LSM6DSL IMU over I2C at 400 kHz. A 256-point CMSIS-DSP FFT over a 156-sample window at 52 Hz separates resting tremor (3-5 Hz), dyskinesia (5-7 Hz), and freezing of gait, with 3-window confirmation to suppress false positives and BLE GATT telemetry out. Interrupt-driven with a polling fallback, so a missed data-ready line degrades the sample rate instead of stalling the device. Thresholds are hand-tuned; there is no on-device ML.

Source →

2D Code Detection and Decoding

CNNPyTorchONNX RuntimeOpenCV

A CNN detector for QR and Data Matrix codes at 96% detection accuracy, reaching up to 100 codes per frame in under 500ms with OpenCV. This is the model behind the recognition work in my C-TEL role: served as a PyTorch detector through ONNX Runtime behind an HTTP service, it decodes at 97% success under harsh plant lighting across 6.5M codes per day, and cut end-to-end recognition time 40%.

BigSMILES Viewer

FastAPICytoscape.jsRDKit

Parses a subset of the BigSMILES polymer notation into a graph, rendering stochastic objects and repeat units as nested boxes instead of flattening them the way a plain SMILES viewer would, with RDKit computing per-unit formula and weight. Code-splitting Cytoscape cut the initial JS bundle 69 percent, from 632 KB to 197 KB. 13 API tests and 7 Playwright specs run in CI, including axe-core WCAG 2.1 AA audits that caught a real contrast failure.

Source →Live →

IICCI Trade Analytics

Django RESTPostgreSQLReact + Recharts

India-Italy bilateral trade analytics platform: Django REST and PostgreSQL behind a React and Recharts dashboard. 21 endpoints covered by 24 tests that pin the aggregation rules, Firebase Google sign-in gated behind a Django-side admin approval step, Excel import in append or replace mode, and short-lived response caching on the reporting routes.

Live →

Studio (client site)

Next.jsSupabaseFramer Motion

Site for a bespoke art-installation studio, built to a Figma design in Next.js, TypeScript, Supabase, and Framer Motion. Behind it sits a custom admin area with a rich-text journal editor, portfolio management, and an enquiry inbox wired to transactional email, so the owner runs the site without a developer. Error boundaries in place and CI running lint and build on every push.

Live →

Notely - Real-Time Notes Platform

DjangoReactJSREST APIs

Built a Django and ReactJS collaboration platform supporting 50+ concurrent users with real-time editing and formula rendering, reducing note-sharing time by 60% for technical coursework.

Scroll for more projects →

contact

I have got just what you need. Lets talk.

+1 (929) 754-7135

chetankodeboyina@gmail.com

New York, USA