>_ nishanth-saka / portfolio

SHEET 00 — TITLE

Nishanth
Saka

Agentic AI · RAG · MLOps · Production Python Systems

ENGINEERNishanth Saka
DISCIPLINEAgentic AI / RAG / MLOps
EXPERIENCE20 YRS
STATUSACTIVELY BUILDING
EMAILnishanth.saka@gmail.com

SHEET 01 — SUMMARY

Always learning. Usually building.

01 / 06

Hands-on AI and software engineer with 20 years of experience designing and shipping production systems across AI, computer vision, real-time collaboration, healthcare, media processing, cloud platforms and public-sector workflows.

Currently building NewsRoom AI, a local-first, multi-service GenAI platform using Python, FastAPI, Ollama and Qwen — with structured generation, deterministic validation, evidence provenance, persistent workflow state, retry and recovery logic, and detailed observability. The platform is being extended into a production-style RAG, MCP and bounded agentic-AI architecture for auditable research over a continuously growing news corpus.

Strong in production Python engineering, API and database design, concurrency-aware backend systems, Docker/Kubernetes, AWS, model serving, evaluation-oriented AI workflows, multimodal pipelines, and translating ambiguous stakeholder needs into resilient, deployable systems.

SHEET 02 — SCHEMATIC

Core modules

02 / 06
U1

Agentic AI & Orchestration

Stateful writer–critic–reviser workflows, bounded retries, model-directed and deterministic routing, schema-validated tool contracts, human-governed actions, LangGraph-oriented research-agent design.

U2

RAG & Context-Aware AI

Canonical document modelling, chunking, embeddings, vector and lexical retrieval, metadata filtering, reranking, context assembly, citations, provenance, abstention, retrieval benchmarks and evaluation.

U3

LLM Engineering

Ollama, Qwen, structured outputs, prompt/context engineering, grounding checks, local LLM deployment, failure taxonomies, revision loops, quality gates and model-provider abstraction.

U4

MCP & Tool Integration

MCP server architecture for typed news search, article retrieval, topic research, source comparison and roundup-evidence tools; separation of MCP contracts from business services.

U5

AI/ML & Computer Vision

YOLO, OpenCV, OCR/ANPR, object tracking, segmentation, depth estimation, Vision Transformers, PyTorch, TensorFlow, InstructBLIP, diffusion pipelines.

U6

MLOps & Model Deployment

FastAPI inference services, GPU-backed deployments, model serving, production diagnostics, monitoring dashboards, CI/CD, API integration.

SHEET 03 — SYSTEMS

Selected production AI systems

03 / 06

NewsRoom AI

IN DEVELOPMENT

MULTI-SERVICE GENAI PLATFORM — PYTHON / FASTAPI / OLLAMA / QWEN

Automates news ingestion, deduplication, story ranking, evidence-grounded script generation, narration, captions, visual selection and short-form video production.

Ingest Dedupe Rank Generate (evidence-grounded) Narrate / Caption Publish (short-form video)

Pravaha AI

LIVE — 2026 → PRESENT

REAL-TIME ANPR & TRAFFIC MONITORING — FASTAPI / GPU / RTSP

Real-time road-safety and traffic-monitoring platform on standard IP cameras: vehicle detection, ANPR/OCR, tracking, violation workflows, analytics, evidence review and live event management for public-sector operations.

RTSP Ingest Detect OCR / ANPR Track Violation Review Operator Dashboard

Fabric Draping AI

SHIPPED — 2025

IMAGE-AI TRANSFORMATION PLATFORM — PYTORCH / OPENCV / VIT

Production image-AI platform for fabric and saree transformation: segmentation, depth estimation, shading, recoloring, seam correction and overlay workflows. Shipped 8+ production modules for fashion-reseller use cases.

Upload Segment Depth Estimate Shade / Recolor Fix Seam Export

SHEET 04 — REVISION HISTORY

Professional experience

04 / 06
Rev Date Role / System Notes
REV 05 Present Principal EngineerNewsRoom AI Multi-service AI platform automating news ingestion, deduplication, story ranking, evidence-grounded script generation, narration, captions, visual selection and short-form video production.
REV 04 2026 – Present Development Head & Principal EngineerPravaha AI — Real-Time ANPR & Traffic Monitoring FastAPI video-processing services for RTSP ingestion, frame processing, OCR, object tracking and GPU-backed deployment. Operator dashboards for live monitoring, evidence review and camera/site rollout across public-sector traffic operations.
REV 03 Jan 2025 – Dec 2025 Principal EngineerFabric Draping AI Fabric-AI platform on FastAPI, React/Vite, Supabase, Railway, Vercel and Hugging Face Spaces. CV pipelines for segmentation, depth estimation, shading, normal-map generation and retinting. Shipped 8+ production modules.
REV 02 Apr 2024 – Dec 2024 Principal EngineerPublic Phone LLC Designed and built Public Phone, a video grievance-collection web app: capture, storage, transcription, translation and publishing. Python, Flask, React, Google Storage, Speech Recognizer, MoviePy, NumPy.
REV 01 Mar 2022 – Apr 2024 Principal EngineerTalview Led development of Code Pair, a multi-participant real-time coding-assessment platform using CRDTs, Webhooks, GraphQL, Supabase, React, Node.js, Docker, Sandpack and CodeMirror.

SHEET 05 — MARGIN NOTES

Education & additional notes

05 / 06

Education

B.E. Electronics & Communication EngineeringCBIT — 2005

Also worth knowing

  • Production AI system architecture
  • Real-time inference and video analytics
  • Public-sector AI deployment workflows
  • AI product prototyping to production
  • Pre-sales, requirements gathering and client communication
  • Engineering leadership and cross-functional delivery
  • Scalable full-stack platform development