01
AI Agent Security & Governance
TrustOS
An independent trust layer for AI agents.
› Before an AI agent acts, TrustOS asks whether it should.
PythonFastAPIGemini APIReactTypeScriptSQLite
Prototype
Available · Ichapore, West Bengal, Indiav1.0 · portfolio.os
I engineer intelligent systems that think, protect and create.
› AI that understands context _
I'm Srinjoy Pramanick (Codexia), a B.Tech CSE student at JIS College of Engineering, passionate about AI/ML, cybersecurity, and full-stack product building.
- Focus
- Trustworthy AI and intelligent systems
- Building
- AI and cybersecurity products
- Location
- Ichapore, West Bengal, India
- Mode
- Learning, building, experimenting
02 · Featured Work
Featured work
Editorial case studies for the systems I've designed and built — each opens as a full deep-dive.
02
Browser Trust & Cybersecurity
CyberTwin AI
A proactive browser trust layer for digital communication.
› Analyzes context before users interact with suspicious content.
TypeScriptReactChrome MV3FastAPIPythonGemini API
Prototype
03
AI Business Intelligence
PulseForge AI
Forging tomorrow's business opportunities today.
› Signals in, opportunities out. Continuously.
PythonFastAPIMongoDB AtlasReactTavilyNewsAPI
View RepositoryPrototype
04
Privacy-First Voice AI
EchoLife
Preserving identity beyond speech loss.
› A privacy-first voice preservation concept.
PythonPyTorchElevenLabs (research)React
Concept
03 · Skills Intelligence Map
Skills intelligence map
No percentage bars. Skills grouped by cluster and confidence, linked to real projects.
Core
- Prompt Engineering
Applied in Projects
- Machine Learning · used in 2 projects
- Generative AI · used in 3 projects
- Agentic AI · used in 1 project
- LLMs · used in 2 projects
- Gemini API · used in 2 projects
- OpenAI API
Currently Exploring
- Computer Vision · next: Vision agents for OCR + reasoning
- TensorFlow
- PyTorch
04 · Engineering Philosophy
Engineering philosophy
I don't build projects only to demonstrate a technology. I build systems around meaningful problems. My work lives at the intersection of AI, cybersecurity, and product engineering — especially systems that must be understandable, trustworthy, and useful in the real world.
- 01Build for real problems
Start with a problem worth solving; the technology is a means, not the point.
- 02Make AI explainable
If a user can't understand a decision, they can't trust it.
- 03Treat security as architecture
Security lives at the boundaries of the system, not as a footer.
- 04Turn prototypes into products
Ship, observe, and iterate with taste. Ideas only matter when they run.
05 · AI Research Lab
AI research lab
Small, honest experiments — what I tried, what happened, what I learned.
Trustworthy AIPrototype
Intent-first agent firewall
- Hypothesis
- Independent intent verification reduces silent agent drift.
- Prototype
- Policy + LLM hybrid evaluator on a mock agent runtime.
- Result
- Catches obvious intent violations; hard cases need scoped rules.
- Lesson
- Rules provide the floor; models provide the ceiling.
- Next
- Add per-agent constitutions.
FastAPIGeminiSQLite
Cybersecurity IntelligenceIn Development
Evidence-first phishing UX
- Hypothesis
- Users trust decisions more when evidence surfaces first.
- Prototype
- Gmail overlay with ranked evidence bundle.
- Result
- Higher perceived trust in preliminary self-tests.
- Lesson
- Explain before you decide.
- Next
- Study behavior on ambiguous messages.
Chrome MV3ReactTypeScript
Business IntelligencePrototype
Business twin from public signals
- Hypothesis
- A living per-company twin beats raw feeds for opportunity discovery.
- Prototype
- Twin schema fed by classified news + hiring signals.
- Result
- Meaningful opportunity ranking on sample industries.
- Lesson
- Provenance is the killer feature.
- Next
- Broaden connectors.
PythonMongoDBTavily
Voice AIExploring
On-device voice identity
- Hypothesis
- Voice identity can be preserved without cloud storage.
- Prototype
- Guided capture + local synthesis research flow.
- Result
- Interaction design validated; runtime is the frontier.
- Lesson
- Consent is a first-class UI element.
- Next
- Evaluate small on-device models.
PyTorchPython
06 · Learning & Building Journey
Learning & building journey
A continuous line — not a resume of years. Where each stage led, and what it taught.
- Stage · 01/ Foundations
Programming & Problem Solving
Learned: Python fundamentals, data structures, and algorithmic thinking.
Lesson · Small consistent problems beat one giant project.
PythonC++Java - Stage · 02/ Web
Frontend & Interfaces
Learned: HTML, CSS, JavaScript, then TypeScript and React.
Built: Component-first UI patterns.
Lesson · Design and engineering share the same craft.
TypeScriptReactTailwind - Stage · 03/ APIs
Backend & Data
Learned: REST APIs, FastAPI, and relational data modelling.
Lesson · The schema is the product boundary.
FastAPIPythonSQL - Stage · 04/ AI/ML
Applied AI & LLMs
Learned: Working with LLM APIs, prompt design, and agent patterns.
Built: First AI-integrated prototypes.
Lesson · Reasoning quality is a design decision, not a model choice.
GeminiOpenAILLMs - Stage · 05/ Security
Cybersecurity & Trust
Learned: Phishing patterns, URL risk, agent governance.
Built: CyberTwin AI browser trust layer.
Lesson · Explainability is the shortest path to user trust.
Chrome MV3TypeScriptFastAPI - Stage · 06/ Hackathons
Product Building Under Pressure
Learned: Scoping, cutting, shipping — then iterating with taste.
Lesson · Constraints reveal the real product.
ReactFastAPIGemini - Stage · 07/ Now
Trustworthy & Agentic AI
Learned: Designing systems that agents cannot silently break.
Built: TrustOS — an AI agent firewall.
Lesson · Governance is a product surface, not a footnote.
FastAPIGeminiPolicy Engines
07 · GitHub & Open Source
Open source
github.com/Codexia-afk
Repositories, prototypes, and experiments. Live activity connects here — no fabricated stats.
Pinned repositories
Connect GitHub data to populate this panel with live stats.
Recent public projects
Connect GitHub data to populate this panel with live stats.
Contribution activity
Connect GitHub data to populate this panel with live stats.
08 · Contact
Let's build intelligent systems people can trust.
I'm open to internships, collaborations, hackathons, AI/ML opportunities, cybersecurity projects, and ambitious product ideas.
email › srinjoypramanick15@gmail.com
linkedin › srinjoy-pramanick
github › @Codexia-afk
bento › bento.me/srinjoy19
resume › Download PDF (June 2026)
status › Open to internships, collaborations, hackathons, and AI/ML opportunities.