Available for full-time roles
Yashwant
Bitra
I build Python backends and the agentic AI systems that run on them.
MS Computer Science · Binghamton University · Open to opportunities across the U.S.
01. About
A bit about me
I'm a Computer Science graduate (MS + BS, Binghamton University) specializing in AI/ML engineering and production system design. My work lives at the intersection of large language models, agentic systems, and real-world infrastructure.
I'm currently the primary engineer on MedKay, a multi-tenant medical imaging platform for a healthcare client, where I also prototype AI-assisted study pre-reads. On the side I build Docs Copilot, a cited document Q&A agent on Amazon Bedrock, and KlevenBook, an open-source GST invoicing app.
Outside of engineering, I'm a competitive chess player and an avid music listener. Currently based in the U.S. and open to full-time opportunities anywhere: on-site, hybrid, or remote.
yashubitra@gmail.com(813) 593-8899United States · Open to Relocation
02. Experience
Where I've Worked
Software Engineer
Cognivance Inc.
May 2026 – Present
Skillman, NJ
- ▸Build MedKay, a multi-tenant medical imaging platform for a healthcare client, as primary engineer on a small team: FastAPI backend with 90 REST operations, Next.js 16 console, database-per-hospital PostgreSQL tenancy, and 568 hermetic pytest tests.
- ▸Integrated dual DICOM archives (Orthanc, dcm4chee) behind a switchable DICOMweb layer after benchmarking 10,861 CT instances, and automated per-hospital provisioning of a PostgreSQL database and Orthanc container via the Docker API.
- ▸Prototyped an AI-assisted study pre-read: a CrewAI flow driving MedGemma on vLLM on a scale-to-zero Modal GPU, with no patient identifiers sent to any model and a guard that blocks marking a finding resolved unless the current read mentions it.
- ▸Design a Secure Imaging Platform proposal for a state government: encrypted ingestion over the company's Secure Transfers product, DICOM de-identification with a separate identity vault, a consent and purpose policy gate, and controlled release to partners.
- ▸Built core components of a prototype LangGraph ReAct agent for the company's file transfer platform: hybrid RAG over a 246-page guide, a LiteLLM gateway A/B testing Groq, Ollama, and OpenAI-compatible backends, Langfuse tracing, and Ragas evals in CI.
- ▸Built ETL pipelines loading 6 compliance frameworks (CMMC, HIPAA, FedRAMP, HITRUST-CSF, DORA, CRI) into one JSONB schema on AWS RDS PostgreSQL, then ran a dry-run-by-default, snapshot-gated cleanup reclaiming storage from 27 inactive organizations across 41 databases.
Research Assistant & Teaching Assistant
State University of New York at Binghamton · School of Computing
RA: May 2025 – May 2026 · TA: August 2024 – May 2025
Binghamton, NY
- ▸Researched LLM architectures, RAG systems, and agentic AI frameworks; delivered reports and presentations on the findings.
- ▸Mentored 40+ students in React, TypeScript, MongoDB, REST API design, and ML pipelines across 10+ capstone projects.
- ▸Automated TypeScript project grading via custom Gradescope scripts, cutting review time by 15 min/submission across 40+ students.
Software Developer / Machine Learning Engineer Intern
CoreCard Software, Inc.
May 2023 – August 2023
Atlanta, GA
- ▸Helped build an internal fraud detection tool: a REST API serving XGBoost and Random Forest models trained with SMOTE on 1M+ transactions, reaching 85% accuracy and flagging 25% more fraud than a baseline model in testing on historical transactions.
- ▸Applied SMOTE to severe class imbalance and engineered features from raw transactions, improving minority-class recall.
- ▸Delivered a 3D analytics dashboard (D3.js, Plotly, Flask) visualizing the same 1M+ historical transactions behind the fraud models.
03. Projects
Things I've Built
- ▸Built a document Q&A agent that streams answers with numbered source citations, choosing per question between hybrid search with reranking over a Bedrock Knowledge Base, a GraphRAG knowledge graph on Neptune Analytics, or a live web browser.
- ▸Exposed the knowledge bases and a Lambda graph-search tool to the agent over MCP through AgentCore Gateway with IAM-signed calls, plus long-term chat memory and per-answer token and cost tracking.
- ▸Chose Mistral Large 3 after Llama 4 Maverick could not call tools while streaming and gpt-oss-120b could not drive the browser.
PythonFastAPINext.jsAmazon Bedrock AgentCoreNeptune AnalyticsLambdaS3MCP
- ▸Built and deployed an open-source GST invoicing app for small medical distributors at book.klevencare.com: tax invoices, quotations, delivery challans, purchase bills, payments, and WhatsApp sharing, with a first live user.
- ▸Wrote PL/pgSQL functions that save each document in one transaction with gap-free per-type numbering, behind row-level security and 9 additive-only migrations.
- ▸Built GSTR-1 sales and purchase CSV exports for GST filing, with received, paid, and balance columns computed from payments.
Next.js 16TypeScriptSupabasePostgreSQLPL/pgSQLTailwind CSSVercel
- ▸Intercepted 25,000+ HTTP/HTTPS flows from a live iOS device via mitmproxy; stored them on LUKS-encrypted volumes.
- ▸Trained an unsupervised deep autoencoder on 21,993 flows; tuning the reconstruction-error threshold cut flagged flows from 1,100 to 103 (90%).
- ▸Shipped a Streamlit real-time traffic-monitoring dashboard replacing manual log triage.
PythonTensorFlowKerasmitmproxyLUKSStreamlit
- ▸Ingested 100K+ posts from Reddit, 4chan, and YouTube into MongoDB with a normalized cross-platform schema.
- ▸Integrated Hugging Face toxicity classifiers at 90% accuracy and built a RAG pipeline (LangChain + MongoDB) for natural-language querying over 100K+ documents.
- ▸Built a Flask dashboard with Chart.js and AJAX visualizing cross-platform trends and toxicity scores.
PythonMongoDBFlaskHugging FaceLangChain
- ▸Ran a full rogue-AP penetration test: deauth via aireplay-ng, an airbase-ng evil twin, and captive-portal credential harvesting.
- ▸Flagged rogue APs in real time by BSSID/MAC mismatch; a pywifi script auto-disconnects clients from untrusted networks.
PythonKali Linuxaircrack-nghostapddnsmasqWireshark
Self-Hosted Home Server
2025 – Present
private
- ▸Operate a 24/7 GPU inference server running Ollama with 7 local LLMs behind Open WebUI, reachable only over a Tailscale mesh with zero open public ports.
- ▸Also host a Samba network share, Immich photo and video backup, and Home Assistant on the same server.
Arch LinuxDockerTailscaleOllamaNVIDIA CUDA
- ▸Co-wrote, in a 4-person team, a proposal and business case for an AI-driven traffic control system that would use city-wide cameras, real-time congestion analysis, and community incident reports to adjust signal timing.
- ▸Planned delivery with a work breakdown structure and Gantt chart, and defined the scope, deliverables, and success measures.
Project ProposalBusiness CaseWork Breakdown StructureGantt ChartSmart CityIoT
04. Skills
Technical Toolkit
Languages
PythonTypeScriptJavaScriptSQLC++C
Backend & APIs
FastAPIFlaskREST API DesignOpenAPIDICOMwebJWT/OAuth2
AI & LLM Systems
LLMsRAGAgentic AILangGraphLangChainLlamaIndexMCPAG-UICrewAI
LLM Tooling & Inference
LiteLLMLangfuseRagasHugging FaceOllamaGroqCerebrasAmazon BedrockvLLMModal
Machine Learning
TensorFlowPyTorchScikit-learnXGBoostSMOTEAutoencoders
Cloud & DevOps
AWS (Lambda, RDS, S3, EC2)DockerGitHub ActionsCI/CDLinux/UnixGit
Databases
PostgreSQLpgvectorSupabaseMongoDBMySQLSQLiteChromaDB
Frontend
ReactNext.jsTailwind CSSCornerstone3D
Certifications
Google Data AnalyticsGoogle CybersecurityRelational Database Systems
05. Education & Leadership
Background
Master of Science in Computer Science
Artificial Intelligence Track
State University of New York at Binghamton
GPA 3.82 / 4.00Dean's List: Fall 2022 – Spring 2024
Bachelor of Science in Computer Science
State University of New York at Binghamton
GPA 3.82 / 4.00Dean's List: Fall 2022 – Spring 2024
Leadership
Vice President
Theta Tau Engineering Fraternity
- ▸Led chapter-wide organizational planning, facilitated leadership meetings, tracked action items, and managed competing stakeholder priorities across concurrent chapter initiatives.
- ▸Served as primary liaison between chapter leadership and membership, resolving conflicts, communicating decisions, and maintaining alignment on chapter goals.
06. Contact
Get In Touch
I'm currently open to full-time opportunities in AI/ML engineering, full-stack development, and data engineering. If you have a role in mind or just want to talk tech, I'd love to hear from you.
yashubitra@gmail.com
linkedin.com/in/yashwant-bitra
GitLab
gitlab.com/yashubitra
Location
Open to U.S. Opportunities