S. Prajwall NarayanaAI Engineer
Open to select engagements

I build LLM systems, and the platform that keeps them up.

Most teams can get a demo working. The hard part is the version that survives real traffic, real data and a bad Tuesday. That is the part I do: retrieval and agent systems with evaluation wired into CI, and the deployment, monitoring and on-call underneath them.

Get in touch See the work GitHub ↗
3 Peer-reviewed papers
10 Shipped projects
12+ Departments using Knotes Central
About

What I build, and how I work

What I build

Multi-agent LLM pipelines for ERP automation over MCP, a code-generation service that turns structured input specs into deployment-ready web applications, and the backend APIs behind them. Long-running, multi-step agent workflows that stay on track without a human babysitting them.

What I run

The platform those systems run on. I established the DevOps/SRE function at Colligence Research from scratch and lead it: CI/CD, infrastructure-as-code with OpenTofu and Ansible on Linode, Docker, Zitadel for SSO and RBAC, observability, alerting and incident runbooks.

Research and publications

Three peer-reviewed papers, one at Springer ICTIS 2025 in Bangkok and two on IEEE Xplore, covering school administration systems, ResNet50 hyperparameter tuning and a web-server controlled rover with object detection. Two more are under review.

Skills

The stack I actually use

Grouped by what it is for, rather than listed as one wall of logos.

AI and LLM systems

Agentic systems · MCP orchestration · Multi-agent workflows · RAG · LangChain · LangGraph · Claude API · LLM integration · Evaluation harnesses

Infrastructure and DevOps

CI/CD · OpenTofu (IaC) · Ansible · Docker · Linux · Linode · Observability and alerting · Incident runbooks · Zitadel (SSO/IAM) · RBAC · Git

Languages and backend

Python · C · C++ · FastAPI · REST APIs · PostgreSQL · SQL · Swift · Rust · Shell scripting

Machine learning and vision

PyTorch · TensorFlow · YOLOv8 · OpenCV · CNN · ResNet · Quantized model deployment

Experience

Where I have shipped

Nov 2025 – present Bengaluru, India

Colligence Research

AI Engineer · Founding team.

  • Works across both halves of the product: the AI and product engineering below, plus end-to-end ownership of the platform it runs on.
  • Established the DevOps/SRE function from scratch and now leads it, CI/CD, infrastructure-as-code with OpenTofu and Ansible on Linode, observability, alerting and incident runbooks.
  • Building a retrieval-augmented system into the infrastructure layer itself, so that infrastructure-as-code and SRE work, runbook lookup, change context, incident response, is grounded in the organisation's own configuration and history rather than hand-carried knowledge. In progress.
  • Replaced manual GitHub deploy workflows, cutting deployment errors for a four-person founding team.
  • Operates the production stack end to end: Docker, Zitadel authentication, monitored deployments with self-healing alerting.
  • Owns system architecture across infrastructure and tooling.

Associate AI Engineer Dec 2025 – Feb 2026

  • Led a small team building the company's first multi-agent LLM pipeline for ERP automation over MCP, keeping long-running multi-step workflows on track without human intervention.
  • Built a code-generation service that parses structured input specs and emits deployment-ready web applications.

Associate AI Engineer Intern Nov 2025 – Dec 2025

  • Prototyped the first LLM integrations and agent workflows.
Oct 2024 – Dec 2024 Remote

RVCE Centre of Excellence

Research Intern, AI and Deep Learning

  • Built a deepfake-detection model from scratch for manipulated-face video, using no external detection APIs.
  • Trained and evaluated on the FaceForensics++ and DFDC datasets.
Aug 2024 – Sep 2024 Bengaluru, India

Ernst & Young India

Associate Consultant Intern, Enterprise Cybersecurity

  • Shadowed senior consultants on cloud-security and identity and access management work.
  • Exposure to the TOGAF and IFTAS frameworks.
Education

Education

Dec 2021 – Aug 2025 Bengaluru, India

R V College of Engineering (VTU)

B.E. Computer Science and Engineering

Projects

Systems I have designed and shipped

Every one is public. The links go to the source.

Jan 2025 – Jun 2025

Arogya-Sathi

Offline-capable multilingual AI healthcare platform

Symptom checking, X-ray fracture detection, prescription summarisation and a wellness chatbot, running fully offline on quantized models, demonstrated on a Raspberry Pi. A deterministic rule-based drug-interaction and crisis-detection safety layer sits underne…

  • Crisis detection: recall 1.00 / precision 0.93
  • Emergency: 1.00 / 1.00
  • Drug screening: 1.00 / 1.00

LangGraph · Ollama · Quantized models · Evaluation harness · CI gate · Raspberry Pi

Aug 2026

Audier

Private reader and audiobook player for Linux, released and installable

A Linux desktop app that reads your own books aloud. It opens PDF, EPUB, Markdown and plain text, reads them in a neural voice with follow-along highlighting, recognises text on scanned pages, and offers optional AI study aids for textbooks across six provider…

  • 202 tests
  • v1.0.0 released, AppImage.deb, tarball
  • Runs fully offline

Flutter · Dart · Piper TTS · Tesseract OCR · GitHub Actions · Snap

agentic-devops

AI-powered CI/CD pipeline generator and monitor

An agent that reads a repository, infers its stack and test setup, and generates a working CI/CD pipeline for it, then watches the runs and explains failures in plain language instead of leaving you to parse red logs. It targets GitHub Actions and GitLab CI fr…

CI/CD · GitHub · GitLab · Agents

pybridge.ai

AI access from a phone

A bridge that puts an LLM behind the channels already on your phone, WhatsApp, Telegram and email, so a model can be driven without an app or a terminal. Each channel is a thin adapter over one routing core that keeps conversation context per user and forwards…

Python · Messaging · LLM integration · WhatsApp · Telegram · Email

Knotes Central

Student resource platform

A resource-sharing platform for students, notes, question papers and course material organised by department, that grew to be used across 12+ departments. It handles upload, categorisation and search so material stays findable instead of scattered across chat …

Web platform · React · Search · File uploads

Deepfake Detection

Multimodal detection on hand-written DSP, calibrated and explainable

A detector that flags AI-synthesised speech and images, built on signal processing written from scratch rather than assembled from libraries: Haar wavelets, DCT-II, STFT, local binary patterns and Benford's-law digit statistics, all hand-implemented in numpy w…

  • AUC 0.957 ±0.015, leave-source-out 5-fold GroupKFold
  • Brier 0.074 calibrated, against 0.264 for the classical baseline
  • Per-artifact recall reported separately, 0.79 to 0.98

numpy · DSP · Wavelets · Calibration · Explainability

Services

What I take on

Scoped as systems, with the tests, instrumentation and runbooks that make them somebody else’s to maintain afterwards.

RAG systems that hold up

Retrieval-augmented systems over your own corpus, hybrid retrieval, reranking, and a labelled evaluation harness wired into CI so quality is measured, not hoped for. The same discipline behind Arogya-Sathi's 62-case, CI-gated safety eval (recall 1.00).

  • Corpus ingestion + hybrid (BM25 + dense) retrieval
  • Cross-encoder reranking
  • Golden-set eval harness as a CI merge gate
  • Cost & latency instrumentation

DevOps / SRE, end to end

The platform your product runs on, CI/CD, infrastructure-as-code, containerisation, observability and incident runbooks. Built the DevOps/SRE function from scratch at a founding-team startup and carry the pager for it.

  • CI/CD pipelines (GitHub Actions, GitLab)
  • IaC with OpenTofu / Ansible
  • Docker & Kubernetes
  • Observability, alerting, runbooks
  • Auth: Zitadel / OAuth / RBAC

LLM & agent pipelines

Multi-agent and LLM systems over MCP with LangGraph/LangChain, kept on-device with local models where the data is sensitive, and shipped as a monitored service, not a demo.

  • Multi-agent workflows (LangGraph, MCP)
  • Local/offline LLM deployment
  • Evaluation & guardrails
  • Backend APIs (FastAPI)
Publications

Peer-reviewed work

Springer ICTIS 2025, Bangkok, ICT for Intelligent Systems, Vol. 9

EduConnect: A Smart Solution for Streamlined School Administrations

Contact

Have something that needs to work in production?

Tell me what you are building and where it is stuck. If I am not the right person for it, I will say so, and usually who is.

prajwallspambox@gmail.com GitHub ↗ Full CV →