Voice-first Lightning wallet
BitVoice Pay
A Bitcoin Lightning wallet you operate entirely through a phone call.
- MIT Bitcoin Hackathon
- 2nd place
Python · Twilio Voice · LNbits · OpenAI · Railway
Boston, MA . open to full-time . available now
Expert Voice AI & RAG Engineer in Boston, MA
Open toApplied AI Engineer, Voice / Speech Engineer, AI Product Engineer, Full-Stack AI Engineer
Voice AI and retrieval engineer in Boston. I build agents that answer real phone calls, RAG systems that cite their sources, and the full-stack product around both - work that calls for a conversational AI engineer who ships in production, not only in a notebook. Recent stacks: Twilio, FastAPI, ChromaDB, Next.js.
specimen 001drag · rotate
11.7k tris · 250 kbmeshopt · webp
- Docs indexed
- 80,000+
- Voice loop budget
- 800 ms
- Retrieval accuracy
- 95%
- Hackathon build
- 48 h
01
Practice · voice AI, RAG, conversational systems
Showcase: Advanced Voice AI and Conversational AI Engineer Projects
Abhishek Sagar Sanda is an applied research engineer based in Boston, specializing in high-performance artificial intelligence systems. His work focuses on dependable, production-ready AI services - especially for early-stage teams where the gap between a prototype and a reliable user-facing product is the entire job. He designs from system constraints and caller experience outward, so each build is both measurable and practical.
BitVoice Pay: A Voice-First Bitcoin Lightning Wallet Solution
BitVoice Pay is a Voice AI product: a Bitcoin Lightning wallet operated entirely by phone call (2nd place, MIT Bitcoin Hackathon). Callers authenticate with a PIN, speak intents, and settle on real LNbits wallets - end to end on Twilio.
Faculty-Sponsored RAG System for Northeastern University
Northeastern University Assistant v2.0 is a faculty-sponsored retrieval system over 80,000+ pages. Answers cite their sources - the accountability bar for any serious RAG engineer deployment at campus scale.
Telephony-Based AI Interview Coaching IVR System
AI Interview Coach IVR runs mock interviews over a live call path, scoring speech rate, filler words, confidence, and clarity - applied telephony for conversational practice.
Further case studies include the AI Dynamic Chatbot (Roli.AI Hackathon winner), the Wyckoff Trading Assistant, and a Manipuri ↔ English Translator trained from scratch for a low-resource pair.
Engage a Professional RAG Engineer for Your Retrieval Systems
System Design Philosophy: Speed, Accuracy, and Reliability
Practice centers on responsive Voice AI and trustworthy retrieval. By designing backward from hard budgets - such as the 800-millisecond response window in voice interactions - he builds architectures that prioritize speed, accuracy, and retention. A system that answers without a verifiable source is a liability; every retrieval stack either cites its material or states that it does not know.
Developing Dependable Conversational AI Engineer Services
As a conversational AI engineer, he ships agents that feel natural on a live phone call rather than only in a demo reel - owning latency budgets, citations, and production failover for teams that need dependable services, not slides. For a focused role page, see AI agent engineer, forward deployed engineer, full-stack AI engineer, ML platform engineer, or book a call.
02
Selected work
Six systems. Every one opens to a full case study.
03
About · how he works
A voice agent has 800 milliseconds. Everything else follows from that.
Past roughly 800 milliseconds of silence, a caller assumes the line dropped. That single number decides the architecture above it: how you chunk, when you stream, what you cache, and which model you can afford to call at all. I design backwards from it rather than discovering it in production.
The same discipline applies to retrieval. A system that returns a confident paragraph with no citation is a liability, not a feature — so everything I ship either returns its source or states plainly that it does not know. The retrieval station further down this page is the live proof: ask it something outside its corpus and watch it refuse.
Boston-based, graduated December 2025, and available now. Most useful to Series A–C teams putting voice agents or retrieval systems in front of real users, where the gap between a demo and a dependable service is the entire job.
04
Trajectory · employment history
Seven roles, from enterprise .NET to a YC voice-agent OS.
Reverse-chronological. Select a role to expand it in place.
Applied AI Engineer · AutoAceYC F25
Jun 2026 - Jul 2026 · Boston, MA
- Built the voice-agent OS for car dealerships: telephony integration carrying live inbound calls end to end.
- Cut voice-loop latency 8.5s → 3.9s by parallelizing call startup, deduplicating graph fetches, and moving blocking TTS init off the async event loop.
- Backfilled 13 dropped calls into recoverable sessions and shipped a three-layer LiveKit fence so concurrent calls fail over instead of dying mid-conversation.
LiveKit · Telephony / SIP · Python · TypeScript · Next.js · Supabase
Education
MS, Information Systems
Northeastern University
Boston, MA · Graduated December 2025
GPA 3.82 / 4.0
B.Tech, Electronics and Communication Engineering
Mahatma Gandhi Institute of Technology
Hyderabad, India
GPA 7.8 / 10
05
Recognition
Top-3 at three MIT hackathons in 14 months.
Every entry below carries its date and, where one exists, an external link you can check.
Awards
Outstanding Master's Student Award - Community Impact
2026One of four university-wide
Northeastern University (MGEN Awards)
Awarded for the Northeastern University Assistant, a faculty-sponsored retrieval system indexing 80,000+ university pages so students get answers with a cited source.
Verify ↗
Hackathon placements
MIT Bitcoin Hackathon - BitVoice Pay
20262nd place
Massachusetts Institute of Technology
A Lightning wallet operated entirely through a phone call, built for people who own a basic phone but not a smartphone. Live dial-in: +1 (888) 805-6555.
Verify ↗MIT × Weights & Biases Multi-Agent Hackathon - Denial Defense
20263rd place
MIT × Weights & Biases
A multi-agent system for contesting insurance claim denials.
Verify ↗Murf.AI Coding Challenge
2025Top-10 finalist
Murf.AI
Built an AI interview coaching IVR on Murf.AI text-to-speech: callers run a mock interview by phone and get scored on speech rate, filler words and clarity.
Verify ↗Roli.AI Hackathon
2024Winner
Roli.AI
Built a dynamic AI chatbot on the Roli.AI platform, wiring conversational state and live model responses into a deployable web client.
Verify ↗
Publication
TensorRT optimization - co-authored paper
2025MIT IMPACT Symposium
Co-authored work on TensorRT inference optimization, presented at the MIT IMPACT Symposium.
No public link available
06
Skills · with proof
Twelve tools, each tied to something shipped.
No proficiency bars. Each entry names what it was used for and links to the work that proves it.
Product surfaces
TypeScript / Next.js
52.1% of public codeProduct surfaces and internal dashboards around AI systems.
AutoAce ticketing →
Models & training
Jupyter / research
20.4% of public codeCorpus assembly, training runs, and BLEU evaluation notebooks.
Manipuri translator →PyTorch
Trains transformers from scratch when no pretrained checkpoint exists.
Manipuri translator →TensorRT
Inference optimization; co-authored work presented at MIT IMPACT 2025.
MIT IMPACT Symposium →
Retrieval & serving
ChromaDB
Vector store behind an 80,000-page retrieval index.
Northeastern Assistant →FastAPI
Serves retrieval answers with confidence scores and source links.
Northeastern Assistant →
Voice & telephony
LiveKit
Three-layer call fence and failover routing under concurrent load.
AutoAce →Twilio Voice
Carries inbound calls into agent logic and back out as speech.
BitVoice Pay →
Backend
Python
12.0% of public codePrimary language for agents, retrieval services, and training code.
BitVoice Pay →.NET / C#
Enterprise services delivered on a Fortune 30 health-insurance engagement.
Elevance Health →
Infrastructure
Postgres / Supabase
Persistence, auth, and row-level isolation for multi-user tools.
AI Interview Coach IVR →Docker / CUDA
GPU-accelerated embedding services with automatic CPU fallback.
Northeastern Assistant →
07
The index · grounded retrieval
Ask about his experience. It cites the file.
Real retrieval, not a chat widget. BM25 over a fixed corpus of resume lines and project READMEs; the answer is composed only from retrieved text. Out-of-corpus questions get refused, not invented.
index livechunks 23p50 127 msqueries 23
corpus: resume.md#summary · resume.md#education · resume.md#contact · resume.md#skills · resume.md#autoace · resume.md#elevance-health
08
Contact
Contact an Applied AI and Voice AI Specialist Today
Toolchain
- Languages
- PythonTypeScriptSQL
- Voice & telephony
- Twilio VoiceMurfAI TTS
- Retrieval & ML
- ChromaDBPyTorchOpenAI / Ollama
- Backend & data
- FastAPINext.jsPostgres / SupabaseDocker / CUDA
Education
MS, Information Systems - Northeastern University, Boston. Graduated December 2025.
Elsewhere
Book a call
Hiring for voice, retrieval, or applied AI product work at a Series A–C team? Send the role and a time window. Replies within one business day.