Remote (India) | AI Agentic Software Engineer · Full-Stack | B.Tech Computer Science Graduate
Building AI agents, agentic workflows, full-stack products, and deterministic evaluation systems — from orchestration and tool calling to verification, backend infrastructure, and production interfaces.
- Repatch (★1) — Autonomous engineering agent: reproduces bugs, authors tests, applies fixes, opens PRs. 7-step state-machine orchestration (UNDERSTAND → EXPLORE → REPRODUCE → PLAN → EXECUTE → VERIFY → SUBMIT). Structured tool calling, Docker/Nixpacks sandboxing, Map of Truth / Semantic Path Grounding.
- vela-cascade-eval — Deterministic evaluation harness for scheduling agents. 500+ seeded scenarios, 713 conflicts generated, 10 machine-checkable invariants, 1,000 determinism checks (byte-identical replay). Naive baseline: 888 double-bookings, 841 priority violations. Cascade-aware agent: zero safety violations, 121 recovery failures (true resource exhaustion).
- groundedness-engine — Zero-LLM structural verification for AI claims. 4 ordered deterministic checks: citation present → section exists → section retrieved → quote matched. Catches "3 years → 5 years" hallucinations without a single model call. Standalone npm package.
- CAREERLENS (★1) — 6-stage semantic resume ↔ JD matching pipeline. SBERT bi-encoder (all-MiniLM-L6-v2) + cosine similarity in 384-dim space. 59-node skill ontology (NetworkX DiGraph) with 2-hop transferability (e.g., Docker → Kubernetes). Implicit skill recovery surfaces 25% more skills vs keyword baselines. 9.1% accuracy gain over tools like Jobscan. Dynamic weight allocation per role type. SHAP-style explainability + PDF report generation.
- Nexa (★1) · Live — AI trip planner powered by Gemini. Personalized itineraries from preferences, not generic guides. React/Next.js, Node/Express, Tailwind, Framer Motion.
| Project | What It Does | Key Signal |
|---|---|---|
| Repatch | Autonomous bug-fixing agent | 7-step state-machine orchestration, structured tool calling, Docker/Nixpacks sandboxing, Map of Truth |
| vela-cascade-eval | Deterministic evaluation harness | 10 invariants, 713 conflicts, seeded replay, honest failure model |
| groundedness-engine | Structural claim verification | Zero-LLM deterministic checks, citation→section→retrieval→quote |
| hermes-agent | Self-improving agent system | Contributor — TUI, MCP, cron, multi-platform gateway |
| Project | What It Does | Key Signal |
|---|---|---|
| vela-cascade-eval | Deterministic evaluation harness | 500+ scenarios, 10 invariants, 1,000 determinism checks |
| groundedness-engine | Zero-LLM structural verification | 4 ordered checks, retrieval tracing, standalone npm pkg |
| Repatch | Empirical TDD loop | Reproduction tests authored before any fix, sandboxed verification |
| Project | What It Does | Stack |
|---|---|---|
| Nexa | AI trip planner — personalized itineraries | Next.js, Gemini API, Tailwind, Framer Motion |
| CAREERLENS | Semantic resume ↔ JD matching | Next.js, FastAPI, SBERT, NetworkX, PostgreSQL |
| EatHere | Location-based food discovery | React 19, Express, MongoDB, Geoapify |
| Job Board | Full-stack MERN job board | React, Node, Express, MongoDB, JWT |
- Building AI agent systems — autonomous engineering agents, orchestration, tool calling, verification, and agentic workflows
- Engineering evaluation-first — deterministic experiments, invariants, seeded scenarios, replayable results, and honest failure models
- Building full-stack products — React/Next.js, Node.js, Python, APIs, databases, deployment, and production interfaces
- Learning in public — exploring how determinism, verification, and reliable failure handling make AI systems trustworthy
- Open to roles — AI engineering, agentic software engineering, full-stack, backend, early-stage startup roles
- Deterministic Agent Evaluation: Why Invariants Beat Vibes
- Zero-LLM Verification: Catching Hallucinations Without Models
- Semantic Resume Matching: Beyond Keyword Overlap
- Repatch — Autonomous engineering agent with empirical TDD loop and sandboxed verification
- vela-cascade-eval — Honest failure model: 121 recovery failures (true resource exhaustion), not weakened assertions
- groundedness-engine — Structural verification layer designed to sit below semantic evaluation
- CAREERLENS — 9.1% accuracy gain over keyword tools via SBERT + ontology transferability
- Finalist — Task API Take-Home Assignment (production-grade MERN app)
"Determinism over vibes. Empirical verification over claims. Honest failure models over hidden edge cases."
I build tools where the verification is the product. If you can't replay it byte-for-byte, it's not evaluated — it's hoped for.
Random Facts
- Run 3-5 agent instances concurrently for eval (Repatch, vela-cascade-eval, groundedness-engine all at once)
- Obsessed with seeded RNG determinism —
mulberry32>Math.random()every time - Powered by filter coffee & late-night invariant debugging
- "It works on my machine" → "It works in the sandbox with Nixpacks detection"
- Will genuinely do a 2-week trial. No bureaucracy. Ship or don't.
if you made it this far, you might as well just reach out