research & benchmarks

things i built, measured, and published.

a benchmark for ai coding agents at brief, plus a run of deep-learning research from before i went full-time on startups. newest first.

01 research & benchmarks
dc
bench

context-augmented code generation: how product context improves ai coding agent decision compliance by 49% new · 2026

drew dillon, kasyap varanasi
arxiv · 2026/paper/code
neurodegenerative disorder detection figure

deep-learning techniques for neurodegenerative disorder detection selected

kasyap varanasi, chandra mohan dasari
eaai · elsevier (if 7.5)/paper/code
fbdnn fetal health figure

fbdnn: fetal health management using a morse-based bi-lstm deep neural network

kasyap varanasi, mure sai jaideep reddy, anupama namburu, senthil kumar mohan, massimiliano ferrara
under review · ieee tpami
genome-sequence prediction figure

neurodegenerative-disorder prediction from curated short mutated genome sequences via hybrid deep neural networks

naman bhatia, kasyap varanasi, parth raghuwanshi, avinash saroj, chandra mohan dasari
icbcb 2024/paper/code
psychnet figure

psychnet: explainable deep neural networks for psychiatric disorders and mental illness

kasyap varanasi, chandra mohan dasari
cict 2022 · ieee/paper
colorectal cancer figure

early detection of colorectal cancer with a patch-based hybrid model and transfer learning selected

kasyap varanasi*, m s jagadeesh*, v s bhagavan, joão manuel r. s. tavares, valentina emilia balas
iccbvic 2022 · springer/paper
more publications
2024
vitadnet: a deep-learning approach for vitamin-d deficiency predictionkasyap varanasi, d. sumathi, m. s. j. reddy, v. s. bhagavan, a. k. cherukuri · j. information & knowledge management
2025
roadsdnet: road-boundary detection with morphological operations and hough transformskasyap varanasi et al. · acm computer vision conference 2025 · −25% navigation error
2023
gastric cancer detection using a hybrid network and shap analysiskasyap varanasi, d. sumathi, k. natarajan · explainable ai for biomedical applications
2022
graph neural networks for aroma prediction from molecular structureskasyap varanasi, v. s. bhagavan, m. s. jagadeesh · ieee gcat 2022
2022
a novel deep-learning framework for diabetic retinopathy detectionkasyap varanasi, c. m. dasari · ieee cict 2022
2022
a novel approach to detect fake news using extreme gradient boostings. s. reddy, s. mandal, kasyap varanasi, a. rk · ieee isdfs 2022
02 learnings, unfiltered

opinions i had to earn the hard way.

  • variational inference: elegant theory, creative excuses in practice.
  • modular networks make you feel like a genius until gradient flow proves otherwise.
  • bayesian optimization is just teaching a computer to pick its own mistakes.
  • sometimes the best fix is still a power cycle, computers and humans alike.
  • “urgent” usually means someone set their own desk on fire.
  • transfer learning is great, until your gpu begs for mercy.
  • explainable ai looks cooler than it is the moment you explain it to a human.
  • edge detection is fun until you realize humans already do it better.
  • “state of the art” usually means “state of constant retraining.”
  • startups don’t run on equity, they run on unpaid overtime and hope.
  • the context window was never the bottleneck. relevance was.
  • most “ai only got me to 80%” gripes are really “i never decided the other 20%.”
  • taste is just compressed experience you can’t quite explain yet.
  • a benchmark you can’t reproduce is just a rumor with a number on it.