Krishna Kankipati

Complete work record

Krishna Kankipati

AI architect and engineer building resilient, interpretable systems that make advanced AI useful in real-world work.

Dallas–Fort Worth, Texas

Krishna KankipatiDownload PDF
LinkedInGitHubPortfolio

Practice

I work at the intersection of large-scale systems engineering and AI research. My goal is to make advanced AI accessible and useful by building products that remain resilient, interpretable, and trustworthy in real-world use.

As a founding AI engineer at SpendRule, I turn complex healthcare agreements into traceable rules for invoice validation. My broader work spans legal AI, healthcare, finance, public sector search, and computer vision. I design the full path from messy source data to a dependable product: tenant-aware intake, grounded retrieval and model outputs, deterministic safeguards, event-driven execution, traces, measurable quality, and workflows that recover safely when something fails.

Hackathon awards

SafeScreen AI 1st place

Qualcomm × Meta ExecuTorch Hackathon · Jun 2026

Team entry ↗

Built with a team: an Android visual safety layer that detects explicit content locally and blurs it across apps. ExecuTorch runs the on-device models; the app does not request internet permission. See the PyTorch winners recap ↗

ARIA 2nd place · prize track

AI Trading Agents Hackathon · Apr 2026

Team entry ↗

Built with team 0TruXt: a multi-strategy trading agent using Kraken paper trading, 13 code-enforced risk gates, and ERC-8004 records for verifiable decisions. Explore the code ↗

Experience

Founding AI Engineer

SpendRule

  • Architected a multi-tenant contract and invoice platform that binds financial decisions to governing source passages, effective dates, facilities, and services.
  • Built SFTP and S3 dropzone intake with archive and file event handlers, deduplication, version-aware document handling, and a content-addressed vault.
  • Designed durable, event-driven extraction, compilation, validation, audit, and recovery stages with checkpoints, retries, and explicit processing state.
  • Separated AI-based semantic matching from deterministic pricing and policy rules; ambiguous cases route to review rather than receiving an unsupported verdict.
  • Added tracing, quality evaluation, asynchronous audit, and guarded dry runs to make the system inspectable and safer to operate.

AI Engineer

108-AI / Casey Legal

  • Built legal AI services spanning document ingestion, OCR, chunking, hybrid retrieval, and grounded single- and multi-document Q&A.
  • Developed stateful contract review and playbook-customization workflows that fetch tenant context, retrieve relevant clauses, review provisions, draft alternatives, validate outputs, and persist results.
  • Wrapped long-running AI work in event-driven execution with retries, status and failure handling; instrumented graph nodes and model calls with tracing for diagnosis and evaluation.
  • Worked on reusable agent interfaces for comparison, redaction, summarization, compliance review, and document exports across containerized Azure services.

AI Developer

OpenView Partners

  • Built predictive analytics and forecasting pipelines in Snowflake using Snowpark and Ray for investment research.
  • Developed a BERT based recommendation engine and natural language to SQL tools for exploring company and warehouse data.
  • Deployed modular LLM applications on GCP with FastAPI, Cloud Run, Pub/Sub, and Cloud Functions.

AI & Web Engineer Intern

NolymitAI

  • Built a multilingual restaurant chatbot with Dialogflow, translation, speech, and OCR services.
  • Deployed a TypeScript and Node.js web interface with Firebase authentication; built additional small-business features including content detection.
  • Coordinated weekly plans and progress reporting across the project team.

Graduate Course Assistant

Stevens Institute of Technology

  • Supported more than 50 students in a graduate algorithms course through instruction and individual guidance.
  • Helped develop and evaluate assignments and examination materials.

Computer Vision Engineer

Vayu Drone Technologies

  • Built geospatial analysis workflows for drone-based land surveys, terrain classification, and environmental monitoring.
  • Trained and deployed image segmentation and object detection models using PyTorch, TensorFlow, and OpenCV.
  • Developed data services with PostgreSQL/PostGIS and AWS for ingestion and analysis of aerial imagery; processed more than 500 images daily.

Consulting work

Investment analysis platform

ExMachina Capital

Finance
  • Architected and built a Spring Boot and AngularJS application for investment and stock analysis, organizational research, deal rooms, and M&A workflows.
  • Added an AI assistant for analysts and helped move the platform from concept to application over eight months.

Intelligent policy search

ALT Solutions · Housing & Urban Development

Public sector
  • Architected search across more than 2,000 mixed format policy documents using AWS Bedrock, OpenSearch, SageMaker, LangChain, Unstructured, and PyMuPDF.
  • Enabled semantic navigation of unstructured documents, improving analyst efficiency by 50%.

Healthcare support copilots

ARS

Healthcare
  • Delivered customized language model APIs for support, appointment scheduling, and recovery center matching.
  • Used SageMaker, Lambda, and API Gateway, with a focus on reliability and handling sensitive information.

System index

The homepage examines four flagship systems in detail. This index includes those systems and the broader project inventory; status describes the work, not a claim of commercial deployment.

Flagship systems

SpendRuleContract intelligence / healthcare

SFTP and dropzone events feed a versioned vault, durable AI validation, source-linked decisions, traces, and audit.

Production work
Casey / 108-AILegal AI platform

Ingestion and hybrid retrieval support traced review, drafting, and validation graphs inside event-driven workflows.

Professional work
SafeScreen AIOn-device visual safety

Local inference and an explicit intervention policy keep screen frames on-device; uncertain deepfake results stay qualified.

Hackathon / 1st place
ARIAAutonomous agent / finance

Code-enforced risk gates separate agent proposals from paper-trade permission; signed records explain accepted and rejected actions.

Hackathon / 2nd in a prize track

Research and product explorations

ChronoscopeHistorical visual search

Exact image crops and OCR chunks become the evaluation unit; human labels anchor search-quality claims.

Active exploration
MITRAIndian Constitution search

Hierarchical, citable source material supports navigation and grounded constitutional Q&A.

Active project
DevCrushTerminal-native social product

Playful developer matching with explicit house-bot labels, opt-in profile data, and a privacy-conscious local scanner.

Active product
Document analysisMultimodal retrieval

Visual embeddings and document Q&A explored with grounded evaluation and offline options for sensitive use.

Active exploration

Interfaces and infrastructure

Excalidraw G.R.I.D.Visual workspace

Diagram organization and AI-assisted edits in an Excalidraw fork.

Public repository
ColNomic embedding APIRetrieval infrastructure

Text and image embedding service with offline loading, adapter support, validation, and health checks.

Public repository
BugFlowSecurity workflow

Report triage with AI classification, semantic duplicate detection, access roles, and background processing.

Public repository
GiffifyAgent tooling / media automation

A portable Codex and Claude Code skill that turns timestamped video clips into optimized GIFs. A deterministic FFmpeg engine handles validation, batch manifests, output planning, and structured results.

Public repository

Browse public repositories

Technical skills

LLM systems

Structured tool outputs, evidence-grounded generation, abstention and review paths, multi-provider integration, RAG, AI agents, MCP, LangGraph, LangChain, LlamaIndex, PaperQA2, evaluation and cost tracing

Model adaptation

SFT, PEFT, RLHF, DPO, PPO and GRPO; Hugging Face, PyTorch, TensorFlow, scikit-learn, experiment tracking with MLflow and W&B

Computer vision

Object detection, segmentation, image classification, multimodal embeddings; YOLO, YOLACT++, Faster R-CNN, UNet, OpenCV, ONNX and ExecuTorch

Data & security

Multi-tenant isolation, versioned records, source provenance, guarded data changes, PostgreSQL/PostGIS, vector search, Snowflake and PySpark

Distributed systems

Durable event-driven execution, idempotency, retries, checkpoints, failure recovery, concurrency control, FastAPI, Inngest and Prefect

Cloud & delivery

AWS (SageMaker, Bedrock, Lambda, S3, OpenSearch), Azure, GCP (Vertex AI, Cloud Run, BigQuery), Docker, Terraform, CI/CD, Langfuse and LangSmith

Application code

Python, TypeScript, React, Next.js, Spring Boot, API design, testing and observability

Education

M.S. in Machine Learning

Stevens Institute of Technology · Hoboken, New Jersey

B.Tech. in Computer Science

VRSEC · Andhra Pradesh, India

Credentials & writing

  • Vision Language Models Bootcamp · OpenCV University, 2025
  • Generative AI with Large Language Models · DeepLearning.AI, 2023
  • LangChain & Vector Databases in Production · Activeloop, 2023
  • Computer Vision Nanodegree · Udacity, 2020

Published an introduction to topological superconductivity and Majorana fermions for quantum engineers, and articles on practical machine learning. See writing on LinkedIn ↗