Skip to content
Now: AI Image Engineer, Precision & Automated Agriculture Lab

Sai Srikar
Reddy Kolli

ML Systems/Evaluation/Agent Tooling

Columbia, Missouri · MS Computer Science, University of Missouri
01

About

I build machine learning systems and the measurement that tells you whether to trust them.

Right now that is infrastructure at the University of Missouri's Precision and Automated Agriculture Lab: I own the training pipeline, the evaluation framework and the data quality tooling. The most useful thing I did there was not a model change. I audited our training data across eight rounds and found 250 mislabelled records, and the ranking of our models changed once they were fixed.

Before graduate school I spent 13 months at MAQ Software, a Microsoft partner, delivering production systems for Fortune 500 clients. Half that role was analysis rather than code: gathering requirements from business stakeholders and writing the specifications developers built from.

Outside coursework I build LLM tooling and ship it. All public, all with tests and CI.

02

Experience

AI Image Engineer

Precision & Automated Agriculture Lab, University of Missouri · Feb 2026 to Present · Columbia, MO
  • Rebuilt the inference path with fp16 mixed precision, channels-last memory layout and batching to reach 14,600 images/sec at 5ms per frame, a 140x gain that took a 500,000 frame corpus from infeasible to routine.
  • Designed a depth-native CNN matching a 15M-parameter ConvNeXt-V2 baseline within two points of macro-F1 using 0.34M parameters, 44x smaller, after benchmarking 9 architectures through one evaluation harness.
  • Built model-assisted tooling to audit training data, surfacing 250 label defects across 8 rounds at 100% precision on a held-out sample.
  • Own the fault-tolerant execution layer: detached Linux orchestration, automatic recovery from transient CUDA faults, and a 56 GB memory-mapped cache that cut per-pass I/O from minutes to seconds.
PyTorchCUDAPythonLinuxRAG

Software Engineer

MAQ Software (Microsoft Partner) · May 2024 to Jun 2025 · Hyderabad, India
  • Systems analysis on the CRM team for Fortune 500 clients: gathered requirement specifications from business stakeholders, analysed their processes, and wrote the specifications and process diagrams developers built from.
  • Architected ETL pipelines in Azure Data Factory ingesting 7 heterogeneous sources into one governed repository, and refactored legacy SQL and PySpark workloads in Databricks, improving data quality 10%.
  • Built semantic models and 10+ Power BI reports with DAX, and cut engineer onboarding 30% with the documentation the team built from.
  • Led a three-person team through the migration from a lakehouse architecture onto Microsoft Fabric.
AzureDatabricksPySparkSQLPower BI

Associate Software Engineer Intern

SP Software (P) Limited · Feb 2024 to Apr 2024 · Hyderabad, India
  • Integrated ESRGAN super-resolution into a production image processing pipeline, improving output clarity 5% while cutting 15ms per image, tuned for detail retention over raw sharpness.
PythonESRGANComputer Vision

Software Developer Intern

HiringFIT · Jul 2023 to Aug 2023 · Remote
  • Shipped the interviewer dashboard and profile pages in React and Next.js across 4 pull requests reviewed and merged by the repository maintainer into a live product.
ReactNext.jsJavaScript
03

Projects

ReproAgent

An agent that reads a paper alongside its public repository and returns an evidence-backed verdict on whether the reported result reproduces. A BLOCKED verdict names the missing artifact rather than failing vaguely, because knowing which claim cannot be checked is the useful output. 20 unit tests, 8 schema-validated outputs, CI.

agent-tab

Cost accounting across three AI coding agents. Claude Code, Codex CLI and Gemini CLI each write a different undocumented session-log format; agent-tab reads all three and renders the day's spend above the shell prompt. No daemon, no database; one cached price fetch you can turn off. TypeScript, tests, CI.

Mizzou Guardian AI

A multi-agent RAG platform, deployed and publicly reachable. Fuses 20,000+ incident records with 2,500+ infrastructure points and returns 0-100 scores that expose the weighted factors behind them, so a user can contest a result instead of being asked to trust it. 36-endpoint FastAPI service on PostgreSQL and PostGIS.

GraphNet-Sentry

Hybrid ResNet-50 and graph neural network for network intrusion detection over a tabular-to-image transform, reaching 99.9% on CICIoT-2023. Then held out entire datasets rather than rows, validating across UNSW-NB15 and TON-IoT to test whether it generalises or only memorises.

DisorderNet

Residue-level classifier predicting intrinsically disordered protein regions from sequence. Hybrid CNN, Transformer and BiLSTM over optional ESM-2 embeddings, split at the protein level rather than the residue level so no sequence appears on both sides. The naive split scores far better and means nothing.

Seat reservation engine

Implemented both concurrency-control strategies rather than picking one, then characterised when each wins: optimistic locking on a version column against distributed Redis locks. Load-tested to 10,000+ concurrent requests with zero double-booking, database reads cut 60%.

Distributed analytics on Spark and Ray

49,191 respondents across 172 variables through a five-stage pipeline, raw CSV to Parquet to feature engineering to Spark SQL, with a Ray-orchestrated Random Forest reaching 0.837 ROC-AUC on the cohort it was built to isolate.

ModelMigrate

Migrates tabular models to Microsoft Fabric semantic models while preserving business-facing names, using a reversible identifier transform plus a change log, and propagating data types across the relationship graph to catch mismatches before they reach the target. FastAPI and React.

04

Skills

Languages

PythonTypeScriptJavaJavaScriptSQLBash

ML & AI

PyTorchCNNsTransformersBiLSTMGraph Neural NetworksESM-2scikit-learnLLMsRAGMulti-agent orchestrationPrompt engineering

Evaluation

Evaluation harness designmacro-F1ROC-AUCMCCBootstrap CIsLeakage-aware splitsCross-dataset generalizationData quality auditing

Backend

FastAPINode.jsReactNext.jsREST APIsPostgreSQLPostGISMySQLRedisApache KafkaConcurrency control

Data & cloud

Apache SparkRayAzureDatabricksMicrosoft FabricPower BIDAXDockerLinuxCI/CDGitHub Actions
05

Education & Credentials

M.S. Computer and Information Sciences

University of Missouri, Columbia · Aug 2025 to May 2027 · 3.95 GPA

B.Tech Computer Science and Engineering

Indian Institute of Information Technology, Sri City · Sep 2020 to Aug 2024 · 7.64 GPA
Certifications

Microsoft Certified: Fabric Analytics Engineer Associate

Microsoft · verify ↗

Road to Practitioner

IBM, via the Mizzou Quantum Innovation Center · invitation-only 12-week quantum computing program with a graded capstone · verify ↗

Neural Networks and Deep Learning

DeepLearning.AI · verify ↗

Supervised Machine Learning: Regression and Classification

DeepLearning.AI and Stanford University · verify ↗
06 · Contact

Let's build something
you can measure.

I'm building in public while I finish my MS at Mizzou. Email is the fastest way to reach me.