Foundation
Platform
Model routing, identity, data, policy, audit, observability, and deployment.
- Gateways
- RBAC
- RAG
- CI/CD
Portfolio / 2026Bengaluru, India
AI Systems
Engineer.
Agentic AI Enterprise AI Infrastructure AI Product Engineering
I build systems where models, tools, data, and people work together: agent orchestration, local-first intelligence, auditable workflows, and complete products from interface to infrastructure.
- Focus
- Agent systems + AI platforms
- Education
- B.E. AI & ML, 2027
- Current mode
- Building in public
00 / SYSTEM MAP
One engineering practice.
Three connected layers.
Coordination
Agents
Supervision, tool use, MCP contracts, human approvals, and evidence trails.
- LangGraph
- MCP
- Tools
- Evaluations
Delivery
Product
Interfaces and workflows that make complex AI systems useful, legible, and safe.
- Next.js
- Flutter
- Firebase
- APIs
01 / SELECTED WORK
Products with
systems behind them.
A focused record of shipped products, field builds, and public systems work. Status labels make the scope and maturity of each project clear.
FEATURED LAUNCHES
Public-facing products and applied builds where product decisions, interfaces, and underlying systems meet.
Shipped work, live testnet, and Android app build
01
Visit
Shipped product
Candidate verification / AI decision support
CREDA
A candidate-verification SaaS that brings structured interview evaluation, repository analysis, and authenticity signals into one review workflow.
Engineering proof
Multi-model fallback, bounded AI spend controls, repository-derived interview prompts, and interaction signals designed to flag low-effort or assisted submissions.
- React
- Node.js
- Supabase
- Redis
- BullMQ
- LLMs
Climate MRV / Algorand
AARNA
A decentralized measurement, reporting, and verification workflow for blue-carbon restoration: project evidence to traceable carbon-credit issuance.
Engineering proof
ARC-4 roles, IPFS evidence, Algorand Standard Asset credits, and an on-chain marketplace model the credit lifecycle end to end.
Built with team BRO CODE at RIFT 2026.
- Algorand
- AlgoPy
- Puya
- React
- TypeScript
- IPFS
03
GitHub
Shipped product
Privacy tech / browser workflow
TrustLayer
A privacy-policy analysis tool that turns dense website terms into a clear, human-readable risk signal.
Engineering proof
An Express-backed analysis layer uses LLM-assisted policy interpretation to surface a visible 1–10 privacy-risk score.
- Chrome extension
- Express
- LLM
- Privacy tech
04
Source
Android companion app
Edge AI / rider safety
SmartHelm
A privacy-first drowsiness-detection system for delivery riders, pairing edge vision and physiological sensing with an Android companion app.
Engineering proof
A Raspberry Pi helmet unit runs MediaPipe face-landmark analysis with MAX30102 heart-rate and SpO2 sensing; Kotlin and Firebase connect rider, device, and fleet view.
- Python
- MediaPipe
- Raspberry Pi
- Kotlin
- Firebase
- Android
02 / AGENTIC + LOCAL SYSTEMS
05
Repository
Enterprise prototype
Agentic operations / procurement integrity
PayGuard-
AgentX
An offline-first retail control-centre prototype for auditable restock, invoice, and fraud-investigation workflows without autonomous payment execution.
Architecture proof
- 01SupervisorLangGraph routing
- 02Agents + critics7 specialists
- 03MCP tools8 schema-bound tools
- 04EvidenceHMAC-signed dossiers
- 05Human decisionApproval queue
06
Repository
Active build
Local-first developer learning
NirmiqCodeSensei
Turns a real repository into a private learning environment with senior-grade review, architecture mapping, code-grounded DSA, and explain-back practice.
Engineering proof
Eight review lenses, repository graphing, MCP integrations, and a localhost SQLite data layer with no telemetry.
- Next.js
- TypeScript
- SQLite
- Drizzle
- MCP
07
Repository
Active MVP
Local academic intelligence
NirmiqResearchOS
A local research workspace for evidence-grounded discovery, synthesis, paper analysis, and exam preparation.
Engineering proof
Hybrid BM25 and vector retrieval with reciprocal-rank fusion, citations, abstention, evaluation flows, and local Ollama inference.
- FastAPI
- Next.js
- BM25
- Chroma
- SQLite
- Ollama
08
Repository
Active capstone
LLM dataset forensics
OriginX-T
A time-aware dataset-forensics system designed to detect contamination risks before they compound into model degradation.
Engineering proof
Models synthetic, recursive, paraphrased, and benchmark-leakage signals using conformal prediction, survival analysis, and SHAP explanations.
- Python
- PyTorch
- Transformers
- PEFT
- SHAP
- MLflow
09
Repository
Product build
Offline-capable product engineering
AscensionOS
A mobile-first personal operating system that converts goals into daily proof, performance memory, and structured weekly review.
Engineering proof
Local cache and offline queue, optional Supabase synchronization, testable domain logic, PWA behavior, and Capacitor Android packaging.
- Next.js
- TypeScript
- Supabase
- Vitest
- Capacitor
- Vercel
10
Repository
Released tool
AI-native developer workflow
Vibecode Project Manager
A cross-platform skill-routing tool for Claude Code that analyzes a project, ranks relevant capabilities, asks for approval, and activates only the selected skills.
Engineering proof
Deterministic lexical ranking, usage-history signals, approval-first configuration writes, Bash and PowerShell support, and CI validation.
- Bash
- PowerShell
- Claude Code
- GitHub Actions
ARCHIVE / EARLIER EVIDENCE
A smaller trail of experiments and applied systems.
02 / ABOUT
Building the connective tissue of useful AI.
My work sits between platform engineering, agent behavior, and product delivery.
I am interested in the parts that make AI dependable beyond a demo: structured tool contracts, human decision boundaries, evidence and audit trails, retrieval quality, local execution, and interfaces that make system state understandable.
The result is a portfolio that spans orchestration and infrastructure without losing sight of the person using the product.
Evidence first
Claims should resolve to source, architecture, tests, or a visible product decision.
Human authority
High-impact workflows need explicit approval boundaries and clear system state.
Local when useful
Privacy, latency, and control can be product features, not only constraints.
03 / CREDENTIALS
Experience
and education.
AI/ML Intern
Equvinoxis Pvt Ltd
Contributed to research and prototyping for AI-assisted facial analysis of pilot stress, working across feature extraction, model experimentation, and human-lab evaluation.
B.E. Artificial Intelligence & Machine Learning
B.N.M. Institute of Technology, Bengaluru
CGPA 8.61 / 10. Coursework and projects across machine learning, deep learning, software engineering, and intelligent systems.
04 / CONTACT
Let’s build the
system behind the AI.
I’m interested in AI systems, agentic AI, platform, applied AI, and AI product engineering opportunities.
siddharthprashoo@gmail.com