$ ls -la ~/projects
projects/
total 7 · 6 shipped · 1 in development
ResearchHQ — Multi-Agent Research Workstation
A general-purpose multi-agent research assistant with two surfaces — a deterministic CLI and a PySide6 desktop studio. Runs on free-tier providers (Groq, Gemini, Ollama) at $0 operational cost.
Multi-Agent Pipeline
- Seven-stage flow: Planner → Searcher → Source Ranker → Extractor → Synthesizer → Verifier → Formatter.
- Each stage emits live progress events, surfaced in both CLI and GUI.
Research Modes & Source Intelligence
- Seven modes (topic, company, competitor, tech, market, news, academic) — each with its own query plan, tier weights, and report schema.
- URLs classified into credibility tiers (Official, Academic, Government, News, Docs, GitHub, …) with mode-specific weighting.
CLI + PySide6 Studio
- Scriptable CLI: `research <mode> "<query>"` with markdown/json/html export.
- Desktop studio: dashboard, live pipeline, history search, side-by-side compare, multi-format export (md/json/html/pdf).
PythonMulti-AgentPySide6GroqGeminiOllamaFastAPIP2P Chat with AES Encryption
Peer-to-peer chat application with real-time AES-encrypted messaging and a SIEM-style monitoring layer for system diagnostics and security.
Encryption & Security
- Server-side AES applied per message; session and key lifecycle managed end-to-end.
- Designed against eavesdropping and replay on the transport layer.
Networking & Sockets
- Python socket programming for true P2P communication without an intermediary.
- Stable handshake and message framing under noisy network conditions.
Operations & Logging
- SIEM-style real-time error log monitoring for system diagnostics.
- Audit trail for connection, auth, and message-flow events.
PythonAES EncryptionSocket ProgrammingSIEMNetwork SecurityNetEngine
Open-source, local-first desktop network toolkit — eight tools in one PyQt6 console: subnet scanner, embedded terminal, multi-tab SSH, SFTP transfer, adapter control, live monitor, API console, and AI automation via Ollama, OpenAI, or Groq.
One Console, Eight Tools
- Subnet scanner, terminal (PowerShell/CMD/WSL/bash), multi-tab SSH, and dual-pane SFTP transfer.
- Adapter control, live ping/latency monitor, and a full API request console — zero tool-hopping.
Local-First AI
- Runs fully offline with Ollama, or plug in OpenAI and Groq — no accounts, no cloud.
- Plain English becomes a reviewed action plan before anything runs; cross-platform native builds.
PythonPyQt6NetworkingSSH / SFTPOllamaAI AutomationAQI Forecasting System
End-to-end air-quality forecasting platform for Delhi with deep-learning predictions, pollutant estimation, and a health-category classifier — served through a modern web UI and REST API.
Modeling Stack
- LSTM model for 24-hour AQI forecasting on time-series data.
- Random Forest pollutant estimator + classifier for health-category labels.
Service Layer
- FastAPI REST endpoints serving predictions to clients.
- React web UI consuming the API for interactive forecasts.
PythonFastAPITensorFlowLSTMRandom ForestReactMovie Recommendation System
ML-driven movie recommendation engine combining content-based, genre-based, and hybrid retrieval — served from a FastAPI backend with a glassmorphism web UI.
Recommendation Engines
- Content-based, genre-based, and hybrid scoring strategies in one engine.
- TF-IDF similarity for fast, interpretable matching.
Delivery
- FastAPI service exposing each engine as a separate endpoint.
- Glassmorphism JavaScript front-end for interactive exploration.
PythonFastAPIScikit-learnTF-IDFJavaScriptHTML/CSSFairPrice
Chrome extension that defeats dynamic pricing by isolating tracking cookies and comparing prices across sessions — exposing hidden fluctuations so users always see a fair, unbiased price.
Cookie Isolation
- Per-session cookie segmentation to prevent vendor fingerprinting.
- Defeats personalized-pricing tracking across visits.
Price Detection
- Cross-session price comparison with real-time UI overlay.
- Highlights hidden price fluctuations directly on the product page.
JavaScriptChrome ExtensionManifest V3Web ScrapingImage Classification with CNNs
Deep-learning image classifier trained on CIFAR-10 with a custom CNN architecture, data augmentation, and dropout regularization — with transfer-learning experiments on ResNet.
Architecture
- Custom CNN with data augmentation and dropout regularization.
- Targets >85% test accuracy on CIFAR-10.
Transfer Learning
- ResNet-backbone experiments for downstream classification tasks.
PythonTensorFlowKerasCNNComputer Vision