EDUCATION
B.Tech in Computer Science & Engineering, VIT Vellore — Aug 2023 to Aug 2027, currently in my final year. Coursework in DSA, OOP, NLP, and AI/ML.
CS @ VIT Vellore · 2023–2027
Sriram Kancherla
ML Engineer · Data Analyst
Final year at VIT Vellore · ex-ML Intern @ FlyRank AI · ex-Academic Intern @ NUS Singapore
Vellore, Tamil Nadu, India
I build and deploy machine learning systems — computer vision, anomaly detection, and applied analytics, with work across finance, healthcare, and cybersecurity.
01——ABOUT
What I work on.
I'm a final-year Computer Science student at VIT Vellore. Most of my work sits between building a model and shipping it — data pipelines, feature engineering, evaluation, and the API or interface that makes the result usable. So far that has covered insider threat detection at NUS, hospital readmission risk, semiconductor defect classification, and an ML-driven stock screener.
FOCUS
ML engineering, data analytics, and finance — with detours into NLP, computer vision, anomaly detection, and full-stack ML apps built on FastAPI and Docker.
GOAL
Work as an ML engineer on systems that run in production — with the pipelines, evaluation, and monitoring that keep them dependable.
LANGUAGES
English, Telugu, and Tamil natively. Hindi and French are a work in progress.
02——EXPERIENCE
Where I've shown up.
Two so far — a June in Singapore and a remote summer. Both beat the syllabus.
JUL 2026 — AUG 2026 · REMOTE · 8 WEEKS
Machine Learning Engineering Intern
FlyRank AI (FlyRank Corp.)
Eight weeks on FlyRank AI's ML Engineering internship, July through August. Real ML engineering the whole way — experimentation, workflow design, and pipelines built to run in production rather than sit in a notebook.
Machine LearningML EngineeringFlyRank AI
JUN 2025 · SINGAPORE
Academic Intern
National University of Singapore (NUS)
Spent June doing exploratory data analysis on large-scale user activity and system log data, looking for the behavioral patterns that turn out to be insider threats. Built and evaluated a detection model using big data analytics and deep learning, then presented the findings under faculty supervision.
EDAInsider Threat DetectionDeep LearningResearch
03——PROJECTS
Things I've built.
Five worth showing. Code for every one of them is on GitHub.
Feb 2026 — Apr 2026 · ML, AI
Image Based Wafer Map Pattern Intelligence
A CNN that looks at silicon wafer maps and tells you what went wrong on the fab floor. The hard part isn't the model — it's wafers arriving at wildly different sizes and a dataset format from another decade. Evaluated properly with confusion matrices, not vibes.
The problem
Wafer maps come off different lines at different die counts, and the public datasets sit in a legacy format. Every map has to be normalised onto a common grid before a model sees it — without flattening the spatial signature that identifies the defect in the first place.
How it works
A CNN over normalised die maps across 8+ labelled defect classes, split 80/20 for train and test. OpenCV handles resizing and augmentation. Evaluation is a per-class confusion matrix rather than one accuracy number, because the classes are badly imbalanced and the rare defects are exactly the ones that cost yield.
Where it stands
Built February to April 2026. Code on GitHub.
8+
Defect Classes
80/20
Train/Test Split
Semiconductor
Domain
PythonPyTorchOpenCVNumPyPandasscikit-learn
Dec 2025 — Present · ML, Full-Stack, Analytics
AInvestify — AI Stock Screener
An AI stock screener that reads the fundamentals and the news at the same time, then calls a stock strong or weak. A Random Forest classifier rates the fundamentals good or bad, an XGBoost regressor turns that into a 0–1 strength score, and VADER handles sentiment across whatever the news is saying that day. Data comes in through yFinance and Google News RSS, models ship out via joblib. The hosted demo is offline while the pipeline is being reworked — the code is on GitHub.
The problem
Fundamentals and sentiment normally get analysed separately, so a screener can call a company cheap on the same day the news says otherwise. Putting both on one scale is the whole problem.
How it works
A Random Forest classifies fundamentals as good or bad; an XGBoost regressor turns that into a 0–1 strength score. VADER scores sentiment over headlines pulled from Google News RSS, fundamentals come in through yFinance, and trained models ship as joblib artefacts behind FastAPI endpoints.
Where it stands
Hosted demo is offline while the pipeline is reworked. Code on GitHub.
4+
ML Models
5
API Endpoints
Rebuilding
Hosted Demo
PythonFastAPIXGBoostRandom ForestVADER NLPyFinanceREST APIsDocker
Sep 2025 — Nov 2025 · AI, Full-Stack
Shiksha Sahayak — Offline-First Education Management Console
An education management platform built for schools where the internet isn't a given. Everything runs locally, including the LLM. Teachers upload material and get worksheets and assessments back; students get a tutor that has actually read the material, thanks to FAISS semantic search. MySQL locally, Firebase for backup, JWT holding the doors.
The problem
Most ed-tech assumes reliable internet and a cloud LLM. Schools without either get nothing at all — which is most of the ones that would benefit the most.
How it works
Everything runs on local infrastructure: a self-hosted LLM, FAISS for semantic search across uploaded material, MySQL for records, JWT for auth. Firebase is a backup target rather than a dependency. Teachers upload material and get worksheets and assessments back; students query a tutor that has actually read it.
Where it stands
Built September to November 2025. Code on GitHub.
JWT Auth
FAISS Search
Local LLM Privacy
FastAPIStreamlitMySQLFAISSJWTFirebase
Sep 2025 — Nov 2025 · ML, Analytics
Healthcare Analytics — IITK D&G Capstone
Predicts whether a patient is coming back within 30 days, using demographics, medical history, admission details, procedures, and discharge outcomes. The point is catching the high-risk cases before the readmission — and the cost — happens. Capstone for the IIT Kanpur E&ICT program.
The problem
Readmission inside 30 days is expensive and often preventable, but the risk is spread thin across demographics, medical history, admission details, procedures and discharge outcomes. No single field predicts it.
How it works
Feature engineering across 6+ data sources into one patient-level view, then supervised classification against a 30-day readmission label — so the output is a ranked risk list a hospital can act on rather than a score in isolation.
Where it stands
Capstone for the IIT Kanpur E&ICT programme. Code and certificate linked above.
30-day
Risk Window
6+
Data Sources
IITK D&G
Partner
Pythonscikit-learnMachine LearningData Analytics
Aug 2025 — Oct 2025 · ML, AI
Insider Threat Detection
User Behavior Analytics on the CERT Insider Threat dataset. Fused five log sources — logon, email, HTTP, file, and USB — into one view, then engineered features for the things people do when they're up to something: late-night USB usage, off-hours logins, non-HTTPS browsing, suspicious email attachments. Autoencoders, Isolation Forest, and LightGBM stacked into one hybrid anomaly pipeline.
The problem
Insider threats look like ordinary activity in any single log. The signal only shows up when logon, email, HTTP, file and USB activity are read as one behavioural timeline per user.
How it works
Five CERT log sources fused into a unified per-user dataset, then 120+ behavioural features engineered on top — late-night USB usage, off-hours logins, non-HTTPS browsing, suspicious attachments. Three models stacked: an autoencoder for reconstruction error, Isolation Forest for density-based outliers, and LightGBM for supervised ranking.
Where it stands
Done during the NUS academic internship. Certificate linked above.
5
Log Sources
3
Models
120+
Features
PythonLightGBMAutoencodersIsolation ForestUBA
04——SKILLS
The stack.
Machine LearningGitEDATableauJavaScriptCSSNumPyFAISSExcelJavaETLGitHubAWS GlueREST APIsOpenCVCPythonData AnalyticsNode.jsHTMLReact NativePandasData ModelingFirebasescikit-learnBig DataC++AWSJoblibSQLStreamlitMATLABPyTorchCI/CDData ManipulationReact.jsDeep LearningLightGBMJWTAzurePostman APIOracle Cloud Infrastructure (OCI)FastAPIComputer VisionMySQLAmazon S3TensorFlowData AcquisitionDockerRandom ForestXGBoostNLP
05——CERTIFICATIONS
Certifications.
The receipts. Newest first.
AWS Certified Cloud Practitioner (CLF-C02)
Amazon Web Services
Aug 2026ViewMicrosoft Azure Fundamentals (AZ-900)
Microsoft
Jul 2026ViewDocker Foundations Professional
Docker
Mar 2026ViewProfessional Certificate in Data Analytics and Generative AI
E&ICT Academy, IIT Kanpur
Dec 2025ViewPostman API Fundamentals Student Expert
Postman
Aug 2025ViewOracle Cloud Infrastructure 2025 Certified Generative AI Professional
Oracle
Jul 2025ViewAWS AI Practitioner Challenge
Amazon Web Services
Jun 2025ViewSupervised Machine Learning: Regression and Classification
DeepLearning.AI · Coursera
May 2025ViewNumPy
CodeChef
Mar 2025ViewPandas
CodeChef
Mar 2025ViewFoundations of GenAI
Udacity
Dec 2024View
06——CONTACT
Let's build something.
Open to full-time roles, research collaborations, or a project that just sounds fun. Based in Vellore, and I reply faster than you'd expect.