Portfolio · 2026

Harshil
Sanghvi

Software engineer working at the intersection of AI, data, and systems. Currently building agent orchestration and conversational AI at Oracle.

Harshil Sanghvi portrait

Based in

Burlington, MA

Est.

2025 —

7

Publications

Journals · preprints · IEEE

3.93

Graduate GPA

Stony Brook / 4.0

9.43

Undergrad GPA

Nirma / 10.0 · Gold

16+

Certifications

Oracle, HackerRank, Coursera

About · 01

Engineer with a researcher's patience and a shipper's bias.

I like problems that sit on the edge of new: complex enough to reward careful thinking, concrete enough to ship something real.

The short version

I build production AI systems at Oracle — agent orchestration, tool platforms, and the knowledge layer that powers enterprise conversational AI.

I hold a Master’s in Computer Science from Stony Brook (3.93 / 4.0) and a Gold Medal undergrad from Nirma University. Seven peer-reviewed publications across AI/ML, security, and neuroscience, four Oracle Cloud certifications, and a lasting obsession with clean abstractions.

Off the keyboard: thrillers, long hikes, and the kind of coffee that resents decaf.

Now
  • Software Engineer

    Oracle

    Jul 2025–Present

  • AI Application Intern (prev.)

    TalentNow

    Jan 2025–May 2025

Education
  • M.S. Computer Science

    Stony Brook University

    3.93 / 4.0 · Aug 2023–May 2025

  • B.Tech. Computer Science & Engineering

    Nirma University

    9.43 / 10. · Gold Medal · Jul 2019–Jun 2023

Daily toolkitFull list ↓
PythonPyTorchLLMs / RAGPostgreSQLJava · SpringOCITypeScript
Say hi

harshil1708@gmail.com

Based

Burlington, MA

Eastern time · open to relocation & remote worldwide

Journey · 02

A compounding résumé.

Seven roles across research labs, trading firms, an HR-tech scale-up, and now Oracle. Each one taught the next.

  • Jul 2025Present
    • Built the orchestration layer that routes requests across conversational agents, data-backed reasoning flows, and external services — tenant-aware auth, endpoint resolution, and streaming SSE that decodes into structured payloads (metadata, UI artifacts, execution context, follow-up actions).
    • Shipped a standards-based tool integration layer — discovery, invocation, schema validation, normalized responses — plus reusable wrappers with structured input contracts, runtime validation, and integration-test coverage.
    • Implemented multi-turn conversation state across the platform: session continuity, follow-up handling, and persistent context across tool calls so multi-step workflows survive instead of resetting per message.
    • Designed the safety layer for AI-driven content changes: semantic guardrails, prompt-injection detection, fail-closed evaluation, and audit logging, plus schema-checked import/export and application/version validation.
    • Built and tuned a production support assistant on Oracle’s conversational AI platform — query preprocessing, response classification, fallback handling, unresolved-request handoff, configurable confidence thresholds, and Slack-channel integration.
    • Designed the knowledge backend that powers the assistant: structured incident-summary generation from raw conversations and images, exact-first retrieval with scored matching and entity extraction, and lifecycle controls (verification, obsolete state, patch, sync) across DB, blob storage, and search index — with batched reads and vectorized status checks for performance.
    PythonFlaskLangGraphLangChainOCI Generative AIMCPOracle DatabaseOracle Digital AssistantSlackpytest
  • Jan 2025May 2025
    • Architected a normalized vector store that cut storage ~37% and similarity comparisons ~99.7%.
    • Built an LLM-powered semantic matcher with adaptive thresholds and a Redis-backed fallback path.
    • Dropped end-to-end matching latency from 60+s to ~13s.
    • Prototyped a LangChain + OpenAI SQL agent that returned results ~50% faster than the existing path.
    LLMsPostgreSQLSQLAlchemyRedisGoElasticsearchAzureKubernetes
  • Aug 2023May 2025
    • Research in 3D diffusion models for medical imaging.
    • Teaching Assistant for Object-Oriented Programming.
  • Feb 2024May 2025
    • Python pipeline parsing .gslog and .ffii files for signaled active avoidance experiments.
    • Turned hours of manual CSV wrangling into seconds, with automated visualization to match.
    • Partnered with Prof. Prerana Shrestha on methodology improvements.
    PythonPandasData VizAutomation
  • Jun 2024Aug 2024
    • Shipped a REST API for Digital Account Management 2.0 that extracts request IDs within a configurable window.
    • Batch applications for AML task assignment and CAT/Tableau reporting.
    • Consolidated 10+ legacy scripts into a unified Spring Boot service.
    JavaSpring BootOracle DBRESTJenkinsTableau
  • Jul 2019Jun 2023
    • Multiple AI/ML internships during undergrad.
    • Published research in AI/ML venues before graduation.
  • Jan 2023Jun 2023
    • Modular Python retrieval service for the Air-CloudPro API — 20% faster data pulls.
    • Combined GESD, Z-Score, Box Plot, Tukey Fences with Isolation Forest, AutoEncoder, LOF.
    • 20% accuracy lift over the legacy rule-based system; 15% reduction in maintenance cost.
    PythonMLIsolation ForestAutoEncoder
  • Jun 2022Aug 2022
    • DistilBERT classifier with 17% accuracy lift and careful EDA-driven preprocessing.
    • FastAPI dashboard showing categories and confidence — 25% less time per email.
    • Embeddings-based cognitive search at 92.5% precision on document/paragraph retrieval.
    DistilBERTFastAPINLPEmbeddings
Skills · 03

Tools I reach for, honestly.

Expert Advanced Intermediate Learning

AI & LLM Engineering

Agents, RAG, and the new LLM stack.

01
  • LLMsExpert
  • RAGExpert
  • LangGraphAdvanced
  • LangChainAdvanced
  • MCP / Tool CallingAdvanced
  • Prompt EngineeringAdvanced

Machine Learning

Deep learning across research and publications.

02
  • PyTorchExpert
  • TensorFlowExpert
  • Deep LearningExpert
  • NLPAdvanced
  • Computer VisionAdvanced
  • scikit-learnAdvanced

Data Science

Stats, modeling, narrative from numbers.

03
  • PandasExpert
  • NumPyExpert
  • StatisticsExpert
  • Feature EngineeringExpert
  • VisualizationAdvanced
  • Time SeriesAdvanced

Languages

The tools I reach for.

04
  • PythonExpert
  • JavaAdvanced
  • SQLAdvanced
  • TypeScriptIntermediate
  • C++Intermediate
  • GoBeginner

Backend

Services and data plumbing.

05
  • FlaskExpert
  • REST APIsExpert
  • Spring BootAdvanced
  • Oracle DatabaseAdvanced
  • PostgreSQLAdvanced
  • MongoDBIntermediate

Infra & Cloud

Ship it, keep it running.

06
  • GitAdvanced
  • OCIIntermediate
  • DockerIntermediate
  • LinuxIntermediate
  • KubernetesBeginner
  • CI/CDIntermediate

Frontend

Enough to build portfolios like this one.

07
  • ReactIntermediate
  • Next.jsIntermediate
  • TypeScriptIntermediate
  • TailwindIntermediate
  • Framer MotionBeginner
  • ResponsiveIntermediate
Selected work · 04

Projects that ship,
or teach me something.

A rotating showcase of systems built across research, ML, and full-stack. The GitHub is the receipt.

01Machine Learning2024

Aviation Delay Prediction

ML pipeline on the DOT flight-delay dataset — geospatial features, gradient boosting, and neural nets — hit 99.2% accuracy.

PythonLightGBMScikit-learnGeospatial
Machine Learning
4 tools
02Full Stack2023

CodeCompanions

Real-time competitive programming platform with live contests, dynamic rankings, and Codeforces API sync.

ReactNode.jsMongoDBSocket.IORedux
Full Stack
5 tools
Also worth a lookAll on GitHub ↗
Research2024

Shrestha Lab Pipeline

Automated data processing for neuroscience research — FreezeFrame and Graphic State experiments.

PythonPandasJinja2
Web2022

QuasarChat

Real-time chat app with auth, emoji, and clean UX — React on the front, Node on the back.

ReactNode.jsSocket.IOMongoDB
Desktop2021

Examify

Desktop examination management system with tracking, attendance, and result reporting.

PythonTkinterMySQL
Research · 05

Peer-reviewed, open, citable.

Seven publications across AI/ML, security, and computational neuroscience — from IEEE conferences to Wiley journals and bioRxiv preprints.

Also reviewed for

Peer review roles across leading AI and ML venues.

Security and Privacy

Wiley

AI security & applied ML

Nov 2025

IEEE Access

IEEE

Applied ML & systems

Jul 2022

Full list, citations, and h-index tracked on Google Scholar.

Google Scholar
Recognition · 06

Receipts, not trophies.

Certifications, research, and academic milestones — the artifacts that prove the work.

Highlight · 2023

Gold Medal

Nirma University

Highlight · 2025

Graduate Excellence

Stony Brook University

Highlight · 2021–2026

7 Research Publications

Wiley · MDPI · IEEE · Springer · bioRxiv

Recommendations

In their own words.

From managers, colleagues, and mentors — via LinkedIn.

I had the pleasure of managing Harshil during his time as an AI Application Intern at TalentNow. From day one, he dove into our AI initiatives with curiosity and drive, quickly mastering new frameworks and languages to which he'd never been exposed before. His thoughtful approach and attention to detail led to marked improvements in both the efficiency and accuracy of our models and pipelines. Beyond his technical acumen, Harshil is a proactive collaborator — eager to take on new challenges, ask the right questions, and share insights that elevate the whole team.

BM

Bobby Mahoney

LinkedIn ↗

VP, Product & Technology

TalentNow

May 2025

I am writing to recommend Harshil, who served as an R&D Intern under my supervision at Johnson Controls Hitachi. We worked closely together on algorithm development. I was constantly amazed not only by his technical skills, but also by his ability to keep the team motivated. Harshil's performance during his internship was exceptional. For any company or team that needs a proven engineer with a get-it-done-with-a-smile approach, Harshil is your guy.

JF

Jingduo Fan

LinkedIn ↗

Senior Engineer

Bosch Home Comfort Group

Jun 2023

Harshil has a really inquisitive, proactive and problem solving mind. During his internship he has done a fabulous work for Northerntrust on the classification and cognitive search model. He has great technical skills and his desire to keep learning is commendable.

AS

Arpita Satpathy

LinkedIn ↗

Vice President

Northern Trust

Aug 2022
Contact · 07

Let's make something
worth shipping.

I read every message. Collaborations, research chats, or just a friendly hello — all welcome.

Send a messageReply within ~24h
Prefer email? →