Srisha KS

AI Engineer · Serverless Infrastructure · LLM Systems

I build AI systems and ship them. I've designed Evidor, a multi-platform document evaluation system on AWS (Lambda, DynamoDB, CloudFront) powered by Gemini with Swiss-system tournament ranking, live at hire.evidor.tech. I've competed in national ML competitions (SIH 2025 winner), published Python tooling to PyPI, and built NLP interfaces over million-profile ocean datasets. Available for remote freelance, contract work, and internships in AI engineering and cloud infrastructure.

Currently available for remote freelance, contract, and internship roles in AI engineering, LLM systems, and AWS serverless infrastructure.

→ What I am building right now

Technical Skills

LanguagesPython, C++, TypeScript, Java, SQL
AI & Data EngineeringPyTorch, TensorFlow, Gradient Boosting, Polars, DuckDB, Transformers, CUDA
Cloud & DevOpsAWS (Lambda, S3, EC2, SQS, CloudFront), Docker, Linux, Git
Backend & DatabasesFastAPI, Flask, DynamoDB, MongoDB
FrontendReact, Next.js, Tailwind CSS

Projects

§1.  High-Constraint Machine Learning

Loss optimization, massive datasets, strict mathematical constraints.

ISRO Air Quality Forecasting[SIH 2025 Winner] — Physics-informed ML pipeline with 200+ features derived from atmospheric chemistry, correcting ISRO's raw CTM forecasts for NO₂ and O₃.
Alpha Radar[2nd Place / 150+ teams] — TCN and Transformer encoders for tick-level Solana momentum, fed into CatBoost under a strict 0.75 recall constraint.
Mallorn Classification — Hierarchical XGBoost pipeline with Hard Negative Mining to detect rare Tidal Disruption Events (5% of data) from astronomical light curves.

§2.  Cloud and Serverless Infrastructure

Backends that hold up under real traffic.

Evidor - AI-Powered Document Evaluation Platform[Live · 3 Platforms · hire.evidor.tech] — Multi-platform document evaluation with Swiss-system tournaments, custom rubrics, and AI-powered ranking. Supports resumes, pitch decks, and assessments across dedicated subdomains.
Blue Intelligence — Natural language interface over 4M+ oceanographic profiles. DuckDB queries an S3 data lake directly, and AI executes analysis code in an AST-validated sandbox.

§3.  Developer Tooling

Tools that make other engineers faster.

IIC Reviews — Fully client-side React SPA for AI code reviews. In-browser key vault with automatic API key rotation handles rate limits without any backend.
QuantFeat — PyPI package implementing range-based volatility estimators (Parkinson, Garman-Klass, Yang-Zhang) and EDA utilities for quant research. (500+ installs)

Experience

ML InternSimpragma Solutions
Nov 2024 - May 2025
  • Built a local LLM inference backend using llama.cpp, replacing a cloud API with small quantized models running on CPU for deterministic, prompt-driven document Q&A with zero external API dependencies.
  • Measured and reduced inference latency significantly by right-sizing models for the task and eliminating network round-trip overhead.
Core Member, Innovation CellInnovation Cell, BMSIT
Aug 2024 - Present
  • Directed nationwide campaigns securing 1000+ registrations for flagship events.
  • Evaluated Round 1 proposals for national prototype competitions.

Certifications

Machine Learning Specialization[view]
Prof. Andrew Ng, Stanford University, Coursera
Deep Learning Specialization[view]
Prof. Andrew Ng, Stanford University, Coursera
Financial Markets
Prof. Robert Shiller, Yale University, Coursera

Education

BMS Institute of Technology
B.Tech in AI & ML2023 - 2027
Coursework: Data Structures, ML, Deep Learning
Get in touchSend a message ↗